]>
Commit | Line | Data |
---|---|---|
2874c5fd | 1 | // SPDX-License-Identifier: GPL-2.0-or-later |
1da177e4 LT |
2 | /* |
3 | * INET An implementation of the TCP/IP protocol suite for the LINUX | |
4 | * operating system. INET is implemented using the BSD Socket | |
5 | * interface as the means of communication with the user level. | |
6 | * | |
7 | * The User Datagram Protocol (UDP). | |
8 | * | |
02c30a84 | 9 | * Authors: Ross Biro |
1da177e4 LT |
10 | * Fred N. van Kempen, <waltje@uWalt.NL.Mugnet.ORG> |
11 | * Arnt Gulbrandsen, <agulbra@nvg.unit.no> | |
113aa838 | 12 | * Alan Cox, <alan@lxorguk.ukuu.org.uk> |
1da177e4 LT |
13 | * Hirokazu Takahashi, <taka@valinux.co.jp> |
14 | * | |
15 | * Fixes: | |
16 | * Alan Cox : verify_area() calls | |
17 | * Alan Cox : stopped close while in use off icmp | |
18 | * messages. Not a fix but a botch that | |
19 | * for udp at least is 'valid'. | |
20 | * Alan Cox : Fixed icmp handling properly | |
21 | * Alan Cox : Correct error for oversized datagrams | |
e905a9ed YH |
22 | * Alan Cox : Tidied select() semantics. |
23 | * Alan Cox : udp_err() fixed properly, also now | |
1da177e4 LT |
24 | * select and read wake correctly on errors |
25 | * Alan Cox : udp_send verify_area moved to avoid mem leak | |
26 | * Alan Cox : UDP can count its memory | |
27 | * Alan Cox : send to an unknown connection causes | |
28 | * an ECONNREFUSED off the icmp, but | |
29 | * does NOT close. | |
30 | * Alan Cox : Switched to new sk_buff handlers. No more backlog! | |
31 | * Alan Cox : Using generic datagram code. Even smaller and the PEEK | |
32 | * bug no longer crashes it. | |
33 | * Fred Van Kempen : Net2e support for sk->broadcast. | |
34 | * Alan Cox : Uses skb_free_datagram | |
35 | * Alan Cox : Added get/set sockopt support. | |
36 | * Alan Cox : Broadcasting without option set returns EACCES. | |
37 | * Alan Cox : No wakeup calls. Instead we now use the callbacks. | |
38 | * Alan Cox : Use ip_tos and ip_ttl | |
39 | * Alan Cox : SNMP Mibs | |
40 | * Alan Cox : MSG_DONTROUTE, and 0.0.0.0 support. | |
41 | * Matt Dillon : UDP length checks. | |
42 | * Alan Cox : Smarter af_inet used properly. | |
43 | * Alan Cox : Use new kernel side addressing. | |
44 | * Alan Cox : Incorrect return on truncated datagram receive. | |
45 | * Arnt Gulbrandsen : New udp_send and stuff | |
46 | * Alan Cox : Cache last socket | |
47 | * Alan Cox : Route cache | |
48 | * Jon Peatfield : Minor efficiency fix to sendto(). | |
49 | * Mike Shaver : RFC1122 checks. | |
50 | * Alan Cox : Nonblocking error fix. | |
51 | * Willy Konynenberg : Transparent proxying support. | |
52 | * Mike McLagan : Routing by source | |
53 | * David S. Miller : New socket lookup architecture. | |
54 | * Last socket cache retained as it | |
55 | * does have a high hit rate. | |
56 | * Olaf Kirch : Don't linearise iovec on sendmsg. | |
57 | * Andi Kleen : Some cleanups, cache destination entry | |
e905a9ed | 58 | * for connect. |
1da177e4 LT |
59 | * Vitaly E. Lavrov : Transparent proxy revived after year coma. |
60 | * Melvin Smith : Check msg_name not msg_namelen in sendto(), | |
61 | * return ENOTCONN for unconnected sockets (POSIX) | |
62 | * Janos Farkas : don't deliver multi/broadcasts to a different | |
63 | * bound-to-device socket | |
64 | * Hirokazu Takahashi : HW checksumming for outgoing UDP | |
65 | * datagrams. | |
66 | * Hirokazu Takahashi : sendfile() on UDP works now. | |
67 | * Arnaldo C. Melo : convert /proc/net/udp to seq_file | |
68 | * YOSHIFUJI Hideaki @USAGI and: Support IPV6_V6ONLY socket option, which | |
69 | * Alexey Kuznetsov: allow both IPv4 and IPv6 sockets to bind | |
70 | * a single port at the same time. | |
71 | * Derek Atkins <derek@ihtfp.com>: Add Encapulation Support | |
342f0234 | 72 | * James Chapman : Add L2TP encapsulation type. |
1da177e4 | 73 | */ |
e905a9ed | 74 | |
afd46503 JP |
75 | #define pr_fmt(fmt) "UDP: " fmt |
76 | ||
7c0f6ba6 | 77 | #include <linux/uaccess.h> |
1da177e4 | 78 | #include <asm/ioctls.h> |
57c8a661 | 79 | #include <linux/memblock.h> |
8203efb3 ED |
80 | #include <linux/highmem.h> |
81 | #include <linux/swap.h> | |
1da177e4 LT |
82 | #include <linux/types.h> |
83 | #include <linux/fcntl.h> | |
84 | #include <linux/module.h> | |
85 | #include <linux/socket.h> | |
86 | #include <linux/sockios.h> | |
14c85021 | 87 | #include <linux/igmp.h> |
6e540309 | 88 | #include <linux/inetdevice.h> |
1da177e4 LT |
89 | #include <linux/in.h> |
90 | #include <linux/errno.h> | |
91 | #include <linux/timer.h> | |
92 | #include <linux/mm.h> | |
1da177e4 | 93 | #include <linux/inet.h> |
1da177e4 | 94 | #include <linux/netdevice.h> |
5a0e3ad6 | 95 | #include <linux/slab.h> |
c752f073 | 96 | #include <net/tcp_states.h> |
1da177e4 LT |
97 | #include <linux/skbuff.h> |
98 | #include <linux/proc_fs.h> | |
99 | #include <linux/seq_file.h> | |
457c4cbc | 100 | #include <net/net_namespace.h> |
1da177e4 | 101 | #include <net/icmp.h> |
421b3885 | 102 | #include <net/inet_hashtables.h> |
e7cc0824 | 103 | #include <net/ip_tunnels.h> |
1da177e4 | 104 | #include <net/route.h> |
1da177e4 LT |
105 | #include <net/checksum.h> |
106 | #include <net/xfrm.h> | |
296f7ea7 | 107 | #include <trace/events/udp.h> |
447167bf | 108 | #include <linux/static_key.h> |
22911fc5 | 109 | #include <trace/events/skb.h> |
076bb0c8 | 110 | #include <net/busy_poll.h> |
ba4e58ec | 111 | #include "udp_impl.h" |
e32ea7e7 | 112 | #include <net/sock_reuseport.h> |
217375a0 | 113 | #include <net/addrconf.h> |
60fb9567 | 114 | #include <net/udp_tunnel.h> |
1da177e4 | 115 | |
f86dcc5a | 116 | struct udp_table udp_table __read_mostly; |
645ca708 | 117 | EXPORT_SYMBOL(udp_table); |
1da177e4 | 118 | |
8d987e5c | 119 | long sysctl_udp_mem[3] __read_mostly; |
95766fff | 120 | EXPORT_SYMBOL(sysctl_udp_mem); |
c482c568 | 121 | |
8d987e5c | 122 | atomic_long_t udp_memory_allocated; |
95766fff HA |
123 | EXPORT_SYMBOL(udp_memory_allocated); |
124 | ||
f86dcc5a ED |
125 | #define MAX_UDP_PORTS 65536 |
126 | #define PORTS_PER_CHAIN (MAX_UDP_PORTS / UDP_HTABLE_SIZE_MIN) | |
98322f22 | 127 | |
63a6fff3 RS |
128 | /* IPCB reference means this can not be used from early demux */ |
129 | static bool udp_lib_exact_dif_match(struct net *net, struct sk_buff *skb) | |
130 | { | |
131 | #if IS_ENABLED(CONFIG_NET_L3_MASTER_DEV) | |
132 | if (!net->ipv4.sysctl_udp_l3mdev_accept && | |
133 | skb && ipv4_l3mdev_skb(IPCB(skb)->flags)) | |
134 | return true; | |
135 | #endif | |
136 | return false; | |
137 | } | |
138 | ||
f24d43c0 | 139 | static int udp_lib_lport_inuse(struct net *net, __u16 num, |
645ca708 | 140 | const struct udp_hslot *hslot, |
98322f22 | 141 | unsigned long *bitmap, |
fe38d2a1 | 142 | struct sock *sk, unsigned int log) |
1da177e4 | 143 | { |
f24d43c0 | 144 | struct sock *sk2; |
ba418fa3 | 145 | kuid_t uid = sock_i_uid(sk); |
25030a7f | 146 | |
ca065d0c | 147 | sk_for_each(sk2, &hslot->head) { |
9d4fb27d JP |
148 | if (net_eq(sock_net(sk2), net) && |
149 | sk2 != sk && | |
d4cada4a | 150 | (bitmap || udp_sk(sk2)->udp_port_hash == num) && |
9d4fb27d JP |
151 | (!sk2->sk_reuse || !sk->sk_reuse) && |
152 | (!sk2->sk_bound_dev_if || !sk->sk_bound_dev_if || | |
153 | sk2->sk_bound_dev_if == sk->sk_bound_dev_if) && | |
fe38d2a1 | 154 | inet_rcv_saddr_equal(sk, sk2, true)) { |
df560056 EG |
155 | if (sk2->sk_reuseport && sk->sk_reuseport && |
156 | !rcu_access_pointer(sk->sk_reuseport_cb) && | |
157 | uid_eq(uid, sock_i_uid(sk2))) { | |
158 | if (!bitmap) | |
159 | return 0; | |
160 | } else { | |
161 | if (!bitmap) | |
162 | return 1; | |
163 | __set_bit(udp_sk(sk2)->udp_port_hash >> log, | |
164 | bitmap); | |
165 | } | |
98322f22 | 166 | } |
4243cdc2 | 167 | } |
25030a7f GR |
168 | return 0; |
169 | } | |
170 | ||
30fff923 ED |
171 | /* |
172 | * Note: we still hold spinlock of primary hash chain, so no other writer | |
173 | * can insert/delete a socket with local_port == num | |
174 | */ | |
175 | static int udp_lib_lport_inuse2(struct net *net, __u16 num, | |
4243cdc2 | 176 | struct udp_hslot *hslot2, |
fe38d2a1 | 177 | struct sock *sk) |
30fff923 ED |
178 | { |
179 | struct sock *sk2; | |
ba418fa3 | 180 | kuid_t uid = sock_i_uid(sk); |
30fff923 ED |
181 | int res = 0; |
182 | ||
183 | spin_lock(&hslot2->lock); | |
ca065d0c | 184 | udp_portaddr_for_each_entry(sk2, &hslot2->head) { |
9d4fb27d JP |
185 | if (net_eq(sock_net(sk2), net) && |
186 | sk2 != sk && | |
187 | (udp_sk(sk2)->udp_port_hash == num) && | |
188 | (!sk2->sk_reuse || !sk->sk_reuse) && | |
189 | (!sk2->sk_bound_dev_if || !sk->sk_bound_dev_if || | |
190 | sk2->sk_bound_dev_if == sk->sk_bound_dev_if) && | |
fe38d2a1 | 191 | inet_rcv_saddr_equal(sk, sk2, true)) { |
df560056 EG |
192 | if (sk2->sk_reuseport && sk->sk_reuseport && |
193 | !rcu_access_pointer(sk->sk_reuseport_cb) && | |
194 | uid_eq(uid, sock_i_uid(sk2))) { | |
195 | res = 0; | |
196 | } else { | |
197 | res = 1; | |
198 | } | |
30fff923 ED |
199 | break; |
200 | } | |
4243cdc2 | 201 | } |
30fff923 ED |
202 | spin_unlock(&hslot2->lock); |
203 | return res; | |
204 | } | |
205 | ||
fe38d2a1 | 206 | static int udp_reuseport_add_sock(struct sock *sk, struct udp_hslot *hslot) |
e32ea7e7 CG |
207 | { |
208 | struct net *net = sock_net(sk); | |
e32ea7e7 CG |
209 | kuid_t uid = sock_i_uid(sk); |
210 | struct sock *sk2; | |
211 | ||
ca065d0c | 212 | sk_for_each(sk2, &hslot->head) { |
e32ea7e7 CG |
213 | if (net_eq(sock_net(sk2), net) && |
214 | sk2 != sk && | |
215 | sk2->sk_family == sk->sk_family && | |
216 | ipv6_only_sock(sk2) == ipv6_only_sock(sk) && | |
217 | (udp_sk(sk2)->udp_port_hash == udp_sk(sk)->udp_port_hash) && | |
218 | (sk2->sk_bound_dev_if == sk->sk_bound_dev_if) && | |
219 | sk2->sk_reuseport && uid_eq(uid, sock_i_uid(sk2)) && | |
fe38d2a1 | 220 | inet_rcv_saddr_equal(sk, sk2, false)) { |
2dbb9b9e MKL |
221 | return reuseport_add_sock(sk, sk2, |
222 | inet_rcv_saddr_any(sk)); | |
e32ea7e7 CG |
223 | } |
224 | } | |
225 | ||
2dbb9b9e | 226 | return reuseport_alloc(sk, inet_rcv_saddr_any(sk)); |
e32ea7e7 CG |
227 | } |
228 | ||
25030a7f | 229 | /** |
6ba5a3c5 | 230 | * udp_lib_get_port - UDP/-Lite port lookup for IPv4 and IPv6 |
25030a7f GR |
231 | * |
232 | * @sk: socket struct in question | |
233 | * @snum: port number to look up | |
25985edc | 234 | * @hash2_nulladdr: AF-dependent hash value in secondary hash chains, |
30fff923 | 235 | * with NULL address |
25030a7f | 236 | */ |
6ba5a3c5 | 237 | int udp_lib_get_port(struct sock *sk, unsigned short snum, |
30fff923 | 238 | unsigned int hash2_nulladdr) |
25030a7f | 239 | { |
512615b6 | 240 | struct udp_hslot *hslot, *hslot2; |
645ca708 | 241 | struct udp_table *udptable = sk->sk_prot->h.udp_table; |
25030a7f | 242 | int error = 1; |
3b1e0a65 | 243 | struct net *net = sock_net(sk); |
1da177e4 | 244 | |
32c1da70 | 245 | if (!snum) { |
9088c560 | 246 | int low, high, remaining; |
95c96174 | 247 | unsigned int rand; |
98322f22 ED |
248 | unsigned short first, last; |
249 | DECLARE_BITMAP(bitmap, PORTS_PER_CHAIN); | |
32c1da70 | 250 | |
0bbf87d8 | 251 | inet_get_local_port_range(net, &low, &high); |
a25de534 | 252 | remaining = (high - low) + 1; |
227b60f5 | 253 | |
63862b5b | 254 | rand = prandom_u32(); |
8fc54f68 | 255 | first = reciprocal_scale(rand, remaining) + low; |
98322f22 ED |
256 | /* |
257 | * force rand to be an odd multiple of UDP_HTABLE_SIZE | |
258 | */ | |
f86dcc5a | 259 | rand = (rand | 1) * (udptable->mask + 1); |
5781b235 ED |
260 | last = first + udptable->mask + 1; |
261 | do { | |
f86dcc5a | 262 | hslot = udp_hashslot(udptable, net, first); |
98322f22 | 263 | bitmap_zero(bitmap, PORTS_PER_CHAIN); |
645ca708 | 264 | spin_lock_bh(&hslot->lock); |
98322f22 | 265 | udp_lib_lport_inuse(net, snum, hslot, bitmap, sk, |
fe38d2a1 | 266 | udptable->log); |
98322f22 ED |
267 | |
268 | snum = first; | |
269 | /* | |
270 | * Iterate on all possible values of snum for this hash. | |
271 | * Using steps of an odd multiple of UDP_HTABLE_SIZE | |
272 | * give us randomization and full range coverage. | |
273 | */ | |
9088c560 | 274 | do { |
98322f22 | 275 | if (low <= snum && snum <= high && |
e3826f1e | 276 | !test_bit(snum >> udptable->log, bitmap) && |
122ff243 | 277 | !inet_is_local_reserved_port(net, snum)) |
98322f22 ED |
278 | goto found; |
279 | snum += rand; | |
280 | } while (snum != first); | |
281 | spin_unlock_bh(&hslot->lock); | |
df560056 | 282 | cond_resched(); |
5781b235 | 283 | } while (++first != last); |
98322f22 | 284 | goto fail; |
645ca708 | 285 | } else { |
f86dcc5a | 286 | hslot = udp_hashslot(udptable, net, snum); |
645ca708 | 287 | spin_lock_bh(&hslot->lock); |
30fff923 ED |
288 | if (hslot->count > 10) { |
289 | int exist; | |
290 | unsigned int slot2 = udp_sk(sk)->udp_portaddr_hash ^ snum; | |
291 | ||
292 | slot2 &= udptable->mask; | |
293 | hash2_nulladdr &= udptable->mask; | |
294 | ||
295 | hslot2 = udp_hashslot2(udptable, slot2); | |
296 | if (hslot->count < hslot2->count) | |
297 | goto scan_primary_hash; | |
298 | ||
fe38d2a1 | 299 | exist = udp_lib_lport_inuse2(net, snum, hslot2, sk); |
30fff923 ED |
300 | if (!exist && (hash2_nulladdr != slot2)) { |
301 | hslot2 = udp_hashslot2(udptable, hash2_nulladdr); | |
302 | exist = udp_lib_lport_inuse2(net, snum, hslot2, | |
fe38d2a1 | 303 | sk); |
30fff923 ED |
304 | } |
305 | if (exist) | |
306 | goto fail_unlock; | |
307 | else | |
308 | goto found; | |
309 | } | |
310 | scan_primary_hash: | |
fe38d2a1 | 311 | if (udp_lib_lport_inuse(net, snum, hslot, NULL, sk, 0)) |
645ca708 ED |
312 | goto fail_unlock; |
313 | } | |
98322f22 | 314 | found: |
c720c7e8 | 315 | inet_sk(sk)->inet_num = snum; |
d4cada4a ED |
316 | udp_sk(sk)->udp_port_hash = snum; |
317 | udp_sk(sk)->udp_portaddr_hash ^= snum; | |
1da177e4 | 318 | if (sk_unhashed(sk)) { |
e32ea7e7 | 319 | if (sk->sk_reuseport && |
fe38d2a1 | 320 | udp_reuseport_add_sock(sk, hslot)) { |
e32ea7e7 CG |
321 | inet_sk(sk)->inet_num = 0; |
322 | udp_sk(sk)->udp_port_hash = 0; | |
323 | udp_sk(sk)->udp_portaddr_hash ^= snum; | |
324 | goto fail_unlock; | |
325 | } | |
326 | ||
ca065d0c | 327 | sk_add_node_rcu(sk, &hslot->head); |
fdcc8aa9 | 328 | hslot->count++; |
c29a0bc4 | 329 | sock_prot_inuse_add(sock_net(sk), sk->sk_prot, 1); |
512615b6 ED |
330 | |
331 | hslot2 = udp_hashslot2(udptable, udp_sk(sk)->udp_portaddr_hash); | |
332 | spin_lock(&hslot2->lock); | |
d894ba18 | 333 | if (IS_ENABLED(CONFIG_IPV6) && sk->sk_reuseport && |
1602f49b DM |
334 | sk->sk_family == AF_INET6) |
335 | hlist_add_tail_rcu(&udp_sk(sk)->udp_portaddr_node, | |
336 | &hslot2->head); | |
d894ba18 | 337 | else |
1602f49b DM |
338 | hlist_add_head_rcu(&udp_sk(sk)->udp_portaddr_node, |
339 | &hslot2->head); | |
512615b6 ED |
340 | hslot2->count++; |
341 | spin_unlock(&hslot2->lock); | |
1da177e4 | 342 | } |
ca065d0c | 343 | sock_set_flag(sk, SOCK_RCU_FREE); |
25030a7f | 344 | error = 0; |
645ca708 ED |
345 | fail_unlock: |
346 | spin_unlock_bh(&hslot->lock); | |
1da177e4 | 347 | fail: |
25030a7f GR |
348 | return error; |
349 | } | |
c482c568 | 350 | EXPORT_SYMBOL(udp_lib_get_port); |
25030a7f | 351 | |
6ba5a3c5 | 352 | int udp_v4_get_port(struct sock *sk, unsigned short snum) |
db8dac20 | 353 | { |
30fff923 | 354 | unsigned int hash2_nulladdr = |
f0b1e64c | 355 | ipv4_portaddr_hash(sock_net(sk), htonl(INADDR_ANY), snum); |
30fff923 | 356 | unsigned int hash2_partial = |
f0b1e64c | 357 | ipv4_portaddr_hash(sock_net(sk), inet_sk(sk)->inet_rcv_saddr, 0); |
30fff923 | 358 | |
d4cada4a | 359 | /* precompute partial secondary hash */ |
30fff923 | 360 | udp_sk(sk)->udp_portaddr_hash = hash2_partial; |
fe38d2a1 | 361 | return udp_lib_get_port(sk, snum, hash2_nulladdr); |
db8dac20 DM |
362 | } |
363 | ||
d1e37288 SX |
364 | static int compute_score(struct sock *sk, struct net *net, |
365 | __be32 saddr, __be16 sport, | |
fb74c277 DA |
366 | __be32 daddr, unsigned short hnum, |
367 | int dif, int sdif, bool exact_dif) | |
645ca708 | 368 | { |
60c04aec JP |
369 | int score; |
370 | struct inet_sock *inet; | |
6da5b0f0 | 371 | bool dev_match; |
645ca708 | 372 | |
60c04aec JP |
373 | if (!net_eq(sock_net(sk), net) || |
374 | udp_sk(sk)->udp_port_hash != hnum || | |
375 | ipv6_only_sock(sk)) | |
376 | return -1; | |
645ca708 | 377 | |
4cdeeee9 PO |
378 | if (sk->sk_rcv_saddr != daddr) |
379 | return -1; | |
60c04aec | 380 | |
4cdeeee9 | 381 | score = (sk->sk_family == PF_INET) ? 2 : 1; |
60c04aec | 382 | |
4cdeeee9 | 383 | inet = inet_sk(sk); |
60c04aec JP |
384 | if (inet->inet_daddr) { |
385 | if (inet->inet_daddr != saddr) | |
386 | return -1; | |
387 | score += 4; | |
388 | } | |
389 | ||
390 | if (inet->inet_dport) { | |
391 | if (inet->inet_dport != sport) | |
392 | return -1; | |
393 | score += 4; | |
394 | } | |
395 | ||
6da5b0f0 MM |
396 | dev_match = udp_sk_bound_dev_eq(net, sk->sk_bound_dev_if, |
397 | dif, sdif); | |
398 | if (!dev_match) | |
399 | return -1; | |
400 | score += 4; | |
fb74c277 | 401 | |
70da268b ED |
402 | if (sk->sk_incoming_cpu == raw_smp_processor_id()) |
403 | score++; | |
645ca708 ED |
404 | return score; |
405 | } | |
406 | ||
6eada011 ED |
407 | static u32 udp_ehashfn(const struct net *net, const __be32 laddr, |
408 | const __u16 lport, const __be32 faddr, | |
409 | const __be16 fport) | |
65cd8033 | 410 | { |
1bbdceef HFS |
411 | static u32 udp_ehash_secret __read_mostly; |
412 | ||
413 | net_get_random_once(&udp_ehash_secret, sizeof(udp_ehash_secret)); | |
414 | ||
65cd8033 | 415 | return __inet_ehashfn(laddr, lport, faddr, fport, |
1bbdceef | 416 | udp_ehash_secret + net_hash_mix(net)); |
65cd8033 HFS |
417 | } |
418 | ||
d1e37288 | 419 | /* called with rcu_read_lock() */ |
5051ebd2 | 420 | static struct sock *udp4_lib_lookup2(struct net *net, |
fb74c277 DA |
421 | __be32 saddr, __be16 sport, |
422 | __be32 daddr, unsigned int hnum, | |
423 | int dif, int sdif, bool exact_dif, | |
424 | struct udp_hslot *hslot2, | |
425 | struct sk_buff *skb) | |
5051ebd2 ED |
426 | { |
427 | struct sock *sk, *result; | |
e94a62f5 | 428 | int score, badness; |
ba418fa3 | 429 | u32 hash = 0; |
5051ebd2 | 430 | |
5051ebd2 | 431 | result = NULL; |
ba418fa3 | 432 | badness = 0; |
ca065d0c | 433 | udp_portaddr_for_each_entry_rcu(sk, &hslot2->head) { |
d1e37288 | 434 | score = compute_score(sk, net, saddr, sport, |
fb74c277 | 435 | daddr, hnum, dif, sdif, exact_dif); |
5051ebd2 | 436 | if (score > badness) { |
e94a62f5 | 437 | if (sk->sk_reuseport) { |
65cd8033 HFS |
438 | hash = udp_ehashfn(net, daddr, hnum, |
439 | saddr, sport); | |
ca065d0c | 440 | result = reuseport_select_sock(sk, hash, skb, |
ed0dfffd | 441 | sizeof(struct udphdr)); |
ca065d0c ED |
442 | if (result) |
443 | return result; | |
ba418fa3 | 444 | } |
ca065d0c ED |
445 | badness = score; |
446 | result = sk; | |
5051ebd2 ED |
447 | } |
448 | } | |
5051ebd2 ED |
449 | return result; |
450 | } | |
451 | ||
db8dac20 DM |
452 | /* UDP is nearly always wildcards out the wazoo, it makes no sense to try |
453 | * harder than this. -DaveM | |
454 | */ | |
fce82338 | 455 | struct sock *__udp4_lib_lookup(struct net *net, __be32 saddr, |
fb74c277 DA |
456 | __be16 sport, __be32 daddr, __be16 dport, int dif, |
457 | int sdif, struct udp_table *udptable, struct sk_buff *skb) | |
db8dac20 | 458 | { |
4cdeeee9 | 459 | struct sock *result; |
db8dac20 | 460 | unsigned short hnum = ntohs(dport); |
4cdeeee9 PO |
461 | unsigned int hash2, slot2; |
462 | struct udp_hslot *hslot2; | |
63a6fff3 | 463 | bool exact_dif = udp_lib_exact_dif_match(net, skb); |
645ca708 | 464 | |
4cdeeee9 PO |
465 | hash2 = ipv4_portaddr_hash(net, daddr, hnum); |
466 | slot2 = hash2 & udptable->mask; | |
467 | hslot2 = &udptable->hash2[slot2]; | |
468 | ||
469 | result = udp4_lib_lookup2(net, saddr, sport, | |
470 | daddr, hnum, dif, sdif, | |
471 | exact_dif, hslot2, skb); | |
472 | if (!result) { | |
473 | hash2 = ipv4_portaddr_hash(net, htonl(INADDR_ANY), hnum); | |
5051ebd2 ED |
474 | slot2 = hash2 & udptable->mask; |
475 | hslot2 = &udptable->hash2[slot2]; | |
5051ebd2 ED |
476 | |
477 | result = udp4_lib_lookup2(net, saddr, sport, | |
4cdeeee9 | 478 | htonl(INADDR_ANY), hnum, dif, sdif, |
63a6fff3 | 479 | exact_dif, hslot2, skb); |
db8dac20 | 480 | } |
4cdeeee9 PO |
481 | if (unlikely(IS_ERR(result))) |
482 | return NULL; | |
db8dac20 DM |
483 | return result; |
484 | } | |
fce82338 | 485 | EXPORT_SYMBOL_GPL(__udp4_lib_lookup); |
db8dac20 | 486 | |
607c4aaf KK |
487 | static inline struct sock *__udp4_lib_lookup_skb(struct sk_buff *skb, |
488 | __be16 sport, __be16 dport, | |
645ca708 | 489 | struct udp_table *udptable) |
607c4aaf KK |
490 | { |
491 | const struct iphdr *iph = ip_hdr(skb); | |
492 | ||
ed7cbbce | 493 | return __udp4_lib_lookup(dev_net(skb->dev), iph->saddr, sport, |
8afdd99a | 494 | iph->daddr, dport, inet_iif(skb), |
fb74c277 | 495 | inet_sdif(skb), udptable, skb); |
607c4aaf KK |
496 | } |
497 | ||
63058308 TH |
498 | struct sock *udp4_lib_lookup_skb(struct sk_buff *skb, |
499 | __be16 sport, __be16 dport) | |
500 | { | |
ed7cbbce | 501 | return __udp4_lib_lookup_skb(skb, sport, dport, &udp_table); |
63058308 TH |
502 | } |
503 | EXPORT_SYMBOL_GPL(udp4_lib_lookup_skb); | |
504 | ||
ca065d0c ED |
505 | /* Must be called under rcu_read_lock(). |
506 | * Does increment socket refcount. | |
507 | */ | |
6e86000c | 508 | #if IS_ENABLED(CONFIG_NF_TPROXY_IPV4) || IS_ENABLED(CONFIG_NF_SOCKET_IPV4) |
bcd41303 KK |
509 | struct sock *udp4_lib_lookup(struct net *net, __be32 saddr, __be16 sport, |
510 | __be32 daddr, __be16 dport, int dif) | |
511 | { | |
ca065d0c ED |
512 | struct sock *sk; |
513 | ||
514 | sk = __udp4_lib_lookup(net, saddr, sport, daddr, dport, | |
fb74c277 | 515 | dif, 0, &udp_table, NULL); |
41c6d650 | 516 | if (sk && !refcount_inc_not_zero(&sk->sk_refcnt)) |
ca065d0c ED |
517 | sk = NULL; |
518 | return sk; | |
bcd41303 KK |
519 | } |
520 | EXPORT_SYMBOL_GPL(udp4_lib_lookup); | |
ca065d0c | 521 | #endif |
bcd41303 | 522 | |
421b3885 SB |
523 | static inline bool __udp_is_mcast_sock(struct net *net, struct sock *sk, |
524 | __be16 loc_port, __be32 loc_addr, | |
525 | __be16 rmt_port, __be32 rmt_addr, | |
fb74c277 | 526 | int dif, int sdif, unsigned short hnum) |
421b3885 SB |
527 | { |
528 | struct inet_sock *inet = inet_sk(sk); | |
529 | ||
530 | if (!net_eq(sock_net(sk), net) || | |
531 | udp_sk(sk)->udp_port_hash != hnum || | |
532 | (inet->inet_daddr && inet->inet_daddr != rmt_addr) || | |
533 | (inet->inet_dport != rmt_port && inet->inet_dport) || | |
534 | (inet->inet_rcv_saddr && inet->inet_rcv_saddr != loc_addr) || | |
535 | ipv6_only_sock(sk) || | |
82ba25c6 | 536 | !udp_sk_bound_dev_eq(net, sk->sk_bound_dev_if, dif, sdif)) |
421b3885 | 537 | return false; |
60d9b031 | 538 | if (!ip_mc_sf_allow(sk, loc_addr, rmt_addr, dif, sdif)) |
421b3885 SB |
539 | return false; |
540 | return true; | |
541 | } | |
542 | ||
a36e185e SB |
543 | DEFINE_STATIC_KEY_FALSE(udp_encap_needed_key); |
544 | void udp_encap_enable(void) | |
545 | { | |
9c480601 | 546 | static_branch_inc(&udp_encap_needed_key); |
a36e185e SB |
547 | } |
548 | EXPORT_SYMBOL(udp_encap_enable); | |
549 | ||
e7cc0824 SB |
550 | /* Handler for tunnels with arbitrary destination ports: no socket lookup, go |
551 | * through error handlers in encapsulations looking for a match. | |
552 | */ | |
553 | static int __udp4_lib_err_encap_no_sk(struct sk_buff *skb, u32 info) | |
554 | { | |
555 | int i; | |
556 | ||
557 | for (i = 0; i < MAX_IPTUN_ENCAP_OPS; i++) { | |
558 | int (*handler)(struct sk_buff *skb, u32 info); | |
92b95364 | 559 | const struct ip_tunnel_encap_ops *encap; |
e7cc0824 | 560 | |
92b95364 PA |
561 | encap = rcu_dereference(iptun_encaps[i]); |
562 | if (!encap) | |
e7cc0824 | 563 | continue; |
92b95364 | 564 | handler = encap->err_handler; |
e7cc0824 SB |
565 | if (handler && !handler(skb, info)) |
566 | return 0; | |
567 | } | |
568 | ||
569 | return -ENOENT; | |
570 | } | |
571 | ||
a36e185e SB |
572 | /* Try to match ICMP errors to UDP tunnels by looking up a socket without |
573 | * reversing source and destination port: this will match tunnels that force the | |
574 | * same destination port on both endpoints (e.g. VXLAN, GENEVE). Note that | |
575 | * lwtunnels might actually break this assumption by being configured with | |
576 | * different destination ports on endpoints, in this case we won't be able to | |
577 | * trace ICMP messages back to them. | |
578 | * | |
e7cc0824 SB |
579 | * If this doesn't match any socket, probe tunnels with arbitrary destination |
580 | * ports (e.g. FoU, GUE): there, the receiving socket is useless, as the port | |
581 | * we've sent packets to won't necessarily match the local destination port. | |
582 | * | |
a36e185e SB |
583 | * Then ask the tunnel implementation to match the error against a valid |
584 | * association. | |
585 | * | |
e7cc0824 SB |
586 | * Return an error if we can't find a match, the socket if we need further |
587 | * processing, zero otherwise. | |
a36e185e SB |
588 | */ |
589 | static struct sock *__udp4_lib_err_encap(struct net *net, | |
590 | const struct iphdr *iph, | |
591 | struct udphdr *uh, | |
592 | struct udp_table *udptable, | |
e7cc0824 | 593 | struct sk_buff *skb, u32 info) |
a36e185e | 594 | { |
a36e185e | 595 | int network_offset, transport_offset; |
a36e185e SB |
596 | struct sock *sk; |
597 | ||
a36e185e SB |
598 | network_offset = skb_network_offset(skb); |
599 | transport_offset = skb_transport_offset(skb); | |
600 | ||
601 | /* Network header needs to point to the outer IPv4 header inside ICMP */ | |
602 | skb_reset_network_header(skb); | |
603 | ||
604 | /* Transport header needs to point to the UDP header */ | |
605 | skb_set_transport_header(skb, iph->ihl << 2); | |
606 | ||
e7cc0824 SB |
607 | sk = __udp4_lib_lookup(net, iph->daddr, uh->source, |
608 | iph->saddr, uh->dest, skb->dev->ifindex, 0, | |
609 | udptable, NULL); | |
610 | if (sk) { | |
611 | int (*lookup)(struct sock *sk, struct sk_buff *skb); | |
612 | struct udp_sock *up = udp_sk(sk); | |
613 | ||
614 | lookup = READ_ONCE(up->encap_err_lookup); | |
615 | if (!lookup || lookup(sk, skb)) | |
616 | sk = NULL; | |
617 | } | |
618 | ||
619 | if (!sk) | |
620 | sk = ERR_PTR(__udp4_lib_err_encap_no_sk(skb, info)); | |
a36e185e SB |
621 | |
622 | skb_set_transport_header(skb, transport_offset); | |
623 | skb_set_network_header(skb, network_offset); | |
624 | ||
625 | return sk; | |
626 | } | |
627 | ||
db8dac20 DM |
628 | /* |
629 | * This routine is called by the ICMP module when it gets some | |
630 | * sort of error condition. If err < 0 then the socket should | |
631 | * be closed and the error returned to the user. If err > 0 | |
632 | * it's just the icmp type << 8 | icmp code. | |
633 | * Header points to the ip header of the error packet. We move | |
634 | * on past this. Then (as it used to claim before adjustment) | |
635 | * header points to the first 8 bytes of the udp header. We need | |
636 | * to find the appropriate port. | |
637 | */ | |
638 | ||
32bbd879 | 639 | int __udp4_lib_err(struct sk_buff *skb, u32 info, struct udp_table *udptable) |
db8dac20 DM |
640 | { |
641 | struct inet_sock *inet; | |
b71d1d42 | 642 | const struct iphdr *iph = (const struct iphdr *)skb->data; |
c482c568 | 643 | struct udphdr *uh = (struct udphdr *)(skb->data+(iph->ihl<<2)); |
db8dac20 DM |
644 | const int type = icmp_hdr(skb)->type; |
645 | const int code = icmp_hdr(skb)->code; | |
a36e185e | 646 | bool tunnel = false; |
db8dac20 DM |
647 | struct sock *sk; |
648 | int harderr; | |
649 | int err; | |
fd54d716 | 650 | struct net *net = dev_net(skb->dev); |
db8dac20 | 651 | |
fd54d716 | 652 | sk = __udp4_lib_lookup(net, iph->daddr, uh->dest, |
f64bf6b8 MM |
653 | iph->saddr, uh->source, skb->dev->ifindex, |
654 | inet_sdif(skb), udptable, NULL); | |
51456b29 | 655 | if (!sk) { |
a36e185e | 656 | /* No socket for error: try tunnels before discarding */ |
e7cc0824 SB |
657 | sk = ERR_PTR(-ENOENT); |
658 | if (static_branch_unlikely(&udp_encap_needed_key)) { | |
659 | sk = __udp4_lib_err_encap(net, iph, uh, udptable, skb, | |
660 | info); | |
661 | if (!sk) | |
662 | return 0; | |
663 | } | |
a36e185e | 664 | |
e7cc0824 | 665 | if (IS_ERR(sk)) { |
a36e185e | 666 | __ICMP_INC_STATS(net, ICMP_MIB_INERRORS); |
e7cc0824 | 667 | return PTR_ERR(sk); |
a36e185e | 668 | } |
e7cc0824 | 669 | |
a36e185e | 670 | tunnel = true; |
db8dac20 DM |
671 | } |
672 | ||
673 | err = 0; | |
674 | harderr = 0; | |
675 | inet = inet_sk(sk); | |
676 | ||
677 | switch (type) { | |
678 | default: | |
679 | case ICMP_TIME_EXCEEDED: | |
680 | err = EHOSTUNREACH; | |
681 | break; | |
682 | case ICMP_SOURCE_QUENCH: | |
683 | goto out; | |
684 | case ICMP_PARAMETERPROB: | |
685 | err = EPROTO; | |
686 | harderr = 1; | |
687 | break; | |
688 | case ICMP_DEST_UNREACH: | |
689 | if (code == ICMP_FRAG_NEEDED) { /* Path MTU discovery */ | |
36393395 | 690 | ipv4_sk_update_pmtu(skb, sk, info); |
db8dac20 DM |
691 | if (inet->pmtudisc != IP_PMTUDISC_DONT) { |
692 | err = EMSGSIZE; | |
693 | harderr = 1; | |
694 | break; | |
695 | } | |
696 | goto out; | |
697 | } | |
698 | err = EHOSTUNREACH; | |
699 | if (code <= NR_ICMP_UNREACH) { | |
700 | harderr = icmp_err_convert[code].fatal; | |
701 | err = icmp_err_convert[code].errno; | |
702 | } | |
703 | break; | |
55be7a9c DM |
704 | case ICMP_REDIRECT: |
705 | ipv4_sk_redirect(skb, sk); | |
1a462d18 | 706 | goto out; |
db8dac20 DM |
707 | } |
708 | ||
709 | /* | |
710 | * RFC1122: OK. Passes ICMP errors back to application, as per | |
711 | * 4.1.3.3. | |
712 | */ | |
a36e185e SB |
713 | if (tunnel) { |
714 | /* ...not for tunnels though: we don't have a sending socket */ | |
715 | goto out; | |
716 | } | |
db8dac20 DM |
717 | if (!inet->recverr) { |
718 | if (!harderr || sk->sk_state != TCP_ESTABLISHED) | |
719 | goto out; | |
b1faf566 | 720 | } else |
c482c568 | 721 | ip_icmp_error(sk, skb, err, uh->dest, info, (u8 *)(uh+1)); |
b1faf566 | 722 | |
db8dac20 DM |
723 | sk->sk_err = err; |
724 | sk->sk_error_report(sk); | |
725 | out: | |
32bbd879 | 726 | return 0; |
db8dac20 DM |
727 | } |
728 | ||
32bbd879 | 729 | int udp_err(struct sk_buff *skb, u32 info) |
db8dac20 | 730 | { |
32bbd879 | 731 | return __udp4_lib_err(skb, info, &udp_table); |
db8dac20 DM |
732 | } |
733 | ||
734 | /* | |
735 | * Throw away all pending data and cancel the corking. Socket is locked. | |
736 | */ | |
36d926b9 | 737 | void udp_flush_pending_frames(struct sock *sk) |
db8dac20 DM |
738 | { |
739 | struct udp_sock *up = udp_sk(sk); | |
740 | ||
741 | if (up->pending) { | |
742 | up->len = 0; | |
743 | up->pending = 0; | |
744 | ip_flush_pending_frames(sk); | |
745 | } | |
746 | } | |
36d926b9 | 747 | EXPORT_SYMBOL(udp_flush_pending_frames); |
db8dac20 DM |
748 | |
749 | /** | |
f6b9664f | 750 | * udp4_hwcsum - handle outgoing HW checksumming |
db8dac20 DM |
751 | * @skb: sk_buff containing the filled-in UDP header |
752 | * (checksum field must be zeroed out) | |
f6b9664f HX |
753 | * @src: source IP address |
754 | * @dst: destination IP address | |
db8dac20 | 755 | */ |
c26bf4a5 | 756 | void udp4_hwcsum(struct sk_buff *skb, __be32 src, __be32 dst) |
db8dac20 | 757 | { |
db8dac20 | 758 | struct udphdr *uh = udp_hdr(skb); |
f6b9664f HX |
759 | int offset = skb_transport_offset(skb); |
760 | int len = skb->len - offset; | |
761 | int hlen = len; | |
db8dac20 DM |
762 | __wsum csum = 0; |
763 | ||
ebbe495f | 764 | if (!skb_has_frag_list(skb)) { |
db8dac20 DM |
765 | /* |
766 | * Only one fragment on the socket. | |
767 | */ | |
768 | skb->csum_start = skb_transport_header(skb) - skb->head; | |
769 | skb->csum_offset = offsetof(struct udphdr, check); | |
f6b9664f HX |
770 | uh->check = ~csum_tcpudp_magic(src, dst, len, |
771 | IPPROTO_UDP, 0); | |
db8dac20 | 772 | } else { |
ebbe495f WC |
773 | struct sk_buff *frags; |
774 | ||
db8dac20 DM |
775 | /* |
776 | * HW-checksum won't work as there are two or more | |
777 | * fragments on the socket so that all csums of sk_buffs | |
778 | * should be together | |
779 | */ | |
ebbe495f | 780 | skb_walk_frags(skb, frags) { |
f6b9664f HX |
781 | csum = csum_add(csum, frags->csum); |
782 | hlen -= frags->len; | |
ebbe495f | 783 | } |
db8dac20 | 784 | |
f6b9664f | 785 | csum = skb_checksum(skb, offset, hlen, csum); |
db8dac20 DM |
786 | skb->ip_summed = CHECKSUM_NONE; |
787 | ||
db8dac20 DM |
788 | uh->check = csum_tcpudp_magic(src, dst, len, IPPROTO_UDP, csum); |
789 | if (uh->check == 0) | |
790 | uh->check = CSUM_MANGLED_0; | |
791 | } | |
792 | } | |
c26bf4a5 | 793 | EXPORT_SYMBOL_GPL(udp4_hwcsum); |
db8dac20 | 794 | |
af5fcba7 TH |
795 | /* Function to set UDP checksum for an IPv4 UDP packet. This is intended |
796 | * for the simple case like when setting the checksum for a UDP tunnel. | |
797 | */ | |
798 | void udp_set_csum(bool nocheck, struct sk_buff *skb, | |
799 | __be32 saddr, __be32 daddr, int len) | |
800 | { | |
801 | struct udphdr *uh = udp_hdr(skb); | |
802 | ||
179bc67f | 803 | if (nocheck) { |
af5fcba7 | 804 | uh->check = 0; |
179bc67f | 805 | } else if (skb_is_gso(skb)) { |
af5fcba7 | 806 | uh->check = ~udp_v4_check(len, saddr, daddr, 0); |
179bc67f EC |
807 | } else if (skb->ip_summed == CHECKSUM_PARTIAL) { |
808 | uh->check = 0; | |
809 | uh->check = udp_v4_check(len, saddr, daddr, lco_csum(skb)); | |
810 | if (uh->check == 0) | |
811 | uh->check = CSUM_MANGLED_0; | |
d75f1306 | 812 | } else { |
af5fcba7 TH |
813 | skb->ip_summed = CHECKSUM_PARTIAL; |
814 | skb->csum_start = skb_transport_header(skb) - skb->head; | |
815 | skb->csum_offset = offsetof(struct udphdr, check); | |
816 | uh->check = ~udp_v4_check(len, saddr, daddr, 0); | |
af5fcba7 TH |
817 | } |
818 | } | |
819 | EXPORT_SYMBOL(udp_set_csum); | |
820 | ||
bec1f6f6 WB |
821 | static int udp_send_skb(struct sk_buff *skb, struct flowi4 *fl4, |
822 | struct inet_cork *cork) | |
db8dac20 | 823 | { |
f6b9664f | 824 | struct sock *sk = skb->sk; |
db8dac20 | 825 | struct inet_sock *inet = inet_sk(sk); |
db8dac20 DM |
826 | struct udphdr *uh; |
827 | int err = 0; | |
828 | int is_udplite = IS_UDPLITE(sk); | |
f6b9664f HX |
829 | int offset = skb_transport_offset(skb); |
830 | int len = skb->len - offset; | |
db8dac20 DM |
831 | __wsum csum = 0; |
832 | ||
db8dac20 DM |
833 | /* |
834 | * Create a UDP header | |
835 | */ | |
836 | uh = udp_hdr(skb); | |
f6b9664f | 837 | uh->source = inet->inet_sport; |
79ab0531 | 838 | uh->dest = fl4->fl4_dport; |
f6b9664f | 839 | uh->len = htons(len); |
db8dac20 DM |
840 | uh->check = 0; |
841 | ||
bec1f6f6 WB |
842 | if (cork->gso_size) { |
843 | const int hlen = skb_network_header_len(skb) + | |
844 | sizeof(struct udphdr); | |
845 | ||
0f149c9f WB |
846 | if (hlen + cork->gso_size > cork->fragsize) { |
847 | kfree_skb(skb); | |
bec1f6f6 | 848 | return -EINVAL; |
0f149c9f WB |
849 | } |
850 | if (skb->len > cork->gso_size * UDP_MAX_SEGMENTS) { | |
851 | kfree_skb(skb); | |
bec1f6f6 | 852 | return -EINVAL; |
0f149c9f WB |
853 | } |
854 | if (sk->sk_no_check_tx) { | |
855 | kfree_skb(skb); | |
a8c744a8 | 856 | return -EINVAL; |
0f149c9f | 857 | } |
ff06342c | 858 | if (skb->ip_summed != CHECKSUM_PARTIAL || is_udplite || |
0f149c9f WB |
859 | dst_xfrm(skb_dst(skb))) { |
860 | kfree_skb(skb); | |
bec1f6f6 | 861 | return -EIO; |
0f149c9f | 862 | } |
bec1f6f6 WB |
863 | |
864 | skb_shinfo(skb)->gso_size = cork->gso_size; | |
865 | skb_shinfo(skb)->gso_type = SKB_GSO_UDP_L4; | |
dfec0ee2 AD |
866 | skb_shinfo(skb)->gso_segs = DIV_ROUND_UP(len - sizeof(uh), |
867 | cork->gso_size); | |
a8c744a8 | 868 | goto csum_partial; |
bec1f6f6 WB |
869 | } |
870 | ||
db8dac20 | 871 | if (is_udplite) /* UDP-Lite */ |
f6b9664f | 872 | csum = udplite_csum(skb); |
db8dac20 | 873 | |
ab2fb7e3 | 874 | else if (sk->sk_no_check_tx) { /* UDP csum off */ |
db8dac20 DM |
875 | |
876 | skb->ip_summed = CHECKSUM_NONE; | |
877 | goto send; | |
878 | ||
879 | } else if (skb->ip_summed == CHECKSUM_PARTIAL) { /* UDP hardware csum */ | |
a8c744a8 | 880 | csum_partial: |
db8dac20 | 881 | |
79ab0531 | 882 | udp4_hwcsum(skb, fl4->saddr, fl4->daddr); |
db8dac20 DM |
883 | goto send; |
884 | ||
f6b9664f HX |
885 | } else |
886 | csum = udp_csum(skb); | |
db8dac20 DM |
887 | |
888 | /* add protocol-dependent pseudo-header */ | |
79ab0531 | 889 | uh->check = csum_tcpudp_magic(fl4->saddr, fl4->daddr, len, |
c482c568 | 890 | sk->sk_protocol, csum); |
db8dac20 DM |
891 | if (uh->check == 0) |
892 | uh->check = CSUM_MANGLED_0; | |
893 | ||
894 | send: | |
b5ec8eea | 895 | err = ip_send_skb(sock_net(sk), skb); |
6ce9e7b5 ED |
896 | if (err) { |
897 | if (err == -ENOBUFS && !inet->recverr) { | |
6aef70a8 ED |
898 | UDP_INC_STATS(sock_net(sk), |
899 | UDP_MIB_SNDBUFERRORS, is_udplite); | |
6ce9e7b5 ED |
900 | err = 0; |
901 | } | |
902 | } else | |
6aef70a8 ED |
903 | UDP_INC_STATS(sock_net(sk), |
904 | UDP_MIB_OUTDATAGRAMS, is_udplite); | |
f6b9664f HX |
905 | return err; |
906 | } | |
907 | ||
908 | /* | |
909 | * Push out all pending data as one UDP datagram. Socket is locked. | |
910 | */ | |
8822b64a | 911 | int udp_push_pending_frames(struct sock *sk) |
f6b9664f HX |
912 | { |
913 | struct udp_sock *up = udp_sk(sk); | |
914 | struct inet_sock *inet = inet_sk(sk); | |
b6f21b26 | 915 | struct flowi4 *fl4 = &inet->cork.fl.u.ip4; |
f6b9664f HX |
916 | struct sk_buff *skb; |
917 | int err = 0; | |
918 | ||
77968b78 | 919 | skb = ip_finish_skb(sk, fl4); |
f6b9664f HX |
920 | if (!skb) |
921 | goto out; | |
922 | ||
bec1f6f6 | 923 | err = udp_send_skb(skb, fl4, &inet->cork.base); |
f6b9664f | 924 | |
db8dac20 DM |
925 | out: |
926 | up->len = 0; | |
927 | up->pending = 0; | |
db8dac20 DM |
928 | return err; |
929 | } | |
8822b64a | 930 | EXPORT_SYMBOL(udp_push_pending_frames); |
db8dac20 | 931 | |
2e8de857 WB |
932 | static int __udp_cmsg_send(struct cmsghdr *cmsg, u16 *gso_size) |
933 | { | |
934 | switch (cmsg->cmsg_type) { | |
935 | case UDP_SEGMENT: | |
936 | if (cmsg->cmsg_len != CMSG_LEN(sizeof(__u16))) | |
937 | return -EINVAL; | |
938 | *gso_size = *(__u16 *)CMSG_DATA(cmsg); | |
939 | return 0; | |
940 | default: | |
941 | return -EINVAL; | |
942 | } | |
943 | } | |
944 | ||
945 | int udp_cmsg_send(struct sock *sk, struct msghdr *msg, u16 *gso_size) | |
946 | { | |
947 | struct cmsghdr *cmsg; | |
948 | bool need_ip = false; | |
949 | int err; | |
950 | ||
951 | for_each_cmsghdr(cmsg, msg) { | |
952 | if (!CMSG_OK(msg, cmsg)) | |
953 | return -EINVAL; | |
954 | ||
955 | if (cmsg->cmsg_level != SOL_UDP) { | |
956 | need_ip = true; | |
957 | continue; | |
958 | } | |
959 | ||
960 | err = __udp_cmsg_send(cmsg, gso_size); | |
961 | if (err) | |
962 | return err; | |
963 | } | |
964 | ||
965 | return need_ip; | |
966 | } | |
967 | EXPORT_SYMBOL_GPL(udp_cmsg_send); | |
968 | ||
1b784140 | 969 | int udp_sendmsg(struct sock *sk, struct msghdr *msg, size_t len) |
db8dac20 DM |
970 | { |
971 | struct inet_sock *inet = inet_sk(sk); | |
972 | struct udp_sock *up = udp_sk(sk); | |
1cedee13 | 973 | DECLARE_SOCKADDR(struct sockaddr_in *, usin, msg->msg_name); |
e474995f | 974 | struct flowi4 fl4_stack; |
b6f21b26 | 975 | struct flowi4 *fl4; |
db8dac20 DM |
976 | int ulen = len; |
977 | struct ipcm_cookie ipc; | |
978 | struct rtable *rt = NULL; | |
979 | int free = 0; | |
980 | int connected = 0; | |
981 | __be32 daddr, faddr, saddr; | |
982 | __be16 dport; | |
983 | u8 tos; | |
984 | int err, is_udplite = IS_UDPLITE(sk); | |
985 | int corkreq = up->corkflag || msg->msg_flags&MSG_MORE; | |
986 | int (*getfrag)(void *, char *, int, int, int, struct sk_buff *); | |
903ab86d | 987 | struct sk_buff *skb; |
f6d8bd05 | 988 | struct ip_options_data opt_copy; |
db8dac20 DM |
989 | |
990 | if (len > 0xFFFF) | |
991 | return -EMSGSIZE; | |
992 | ||
993 | /* | |
994 | * Check the flags. | |
995 | */ | |
996 | ||
c482c568 | 997 | if (msg->msg_flags & MSG_OOB) /* Mirror BSD error message compatibility */ |
db8dac20 DM |
998 | return -EOPNOTSUPP; |
999 | ||
903ab86d HX |
1000 | getfrag = is_udplite ? udplite_getfrag : ip_generic_getfrag; |
1001 | ||
f5fca608 | 1002 | fl4 = &inet->cork.fl.u.ip4; |
db8dac20 DM |
1003 | if (up->pending) { |
1004 | /* | |
1005 | * There are pending frames. | |
1006 | * The socket lock must be held while it's corked. | |
1007 | */ | |
1008 | lock_sock(sk); | |
1009 | if (likely(up->pending)) { | |
1010 | if (unlikely(up->pending != AF_INET)) { | |
1011 | release_sock(sk); | |
1012 | return -EINVAL; | |
1013 | } | |
1014 | goto do_append_data; | |
1015 | } | |
1016 | release_sock(sk); | |
1017 | } | |
1018 | ulen += sizeof(struct udphdr); | |
1019 | ||
1020 | /* | |
1021 | * Get and verify the address. | |
1022 | */ | |
1cedee13 | 1023 | if (usin) { |
db8dac20 DM |
1024 | if (msg->msg_namelen < sizeof(*usin)) |
1025 | return -EINVAL; | |
1026 | if (usin->sin_family != AF_INET) { | |
1027 | if (usin->sin_family != AF_UNSPEC) | |
1028 | return -EAFNOSUPPORT; | |
1029 | } | |
1030 | ||
1031 | daddr = usin->sin_addr.s_addr; | |
1032 | dport = usin->sin_port; | |
1033 | if (dport == 0) | |
1034 | return -EINVAL; | |
1035 | } else { | |
1036 | if (sk->sk_state != TCP_ESTABLISHED) | |
1037 | return -EDESTADDRREQ; | |
c720c7e8 ED |
1038 | daddr = inet->inet_daddr; |
1039 | dport = inet->inet_dport; | |
db8dac20 DM |
1040 | /* Open fast path for connected socket. |
1041 | Route will not be used, if at least one option is set. | |
1042 | */ | |
1043 | connected = 1; | |
1044 | } | |
db8dac20 | 1045 | |
35178206 | 1046 | ipcm_init_sk(&ipc, inet); |
bec1f6f6 | 1047 | ipc.gso_size = up->gso_size; |
bf84a010 | 1048 | |
db8dac20 | 1049 | if (msg->msg_controllen) { |
2e8de857 WB |
1050 | err = udp_cmsg_send(sk, msg, &ipc.gso_size); |
1051 | if (err > 0) | |
1052 | err = ip_cmsg_send(sk, msg, &ipc, | |
1053 | sk->sk_family == AF_INET6); | |
1054 | if (unlikely(err < 0)) { | |
91948309 | 1055 | kfree(ipc.opt); |
db8dac20 | 1056 | return err; |
91948309 | 1057 | } |
db8dac20 DM |
1058 | if (ipc.opt) |
1059 | free = 1; | |
1060 | connected = 0; | |
1061 | } | |
f6d8bd05 ED |
1062 | if (!ipc.opt) { |
1063 | struct ip_options_rcu *inet_opt; | |
1064 | ||
1065 | rcu_read_lock(); | |
1066 | inet_opt = rcu_dereference(inet->inet_opt); | |
1067 | if (inet_opt) { | |
1068 | memcpy(&opt_copy, inet_opt, | |
1069 | sizeof(*inet_opt) + inet_opt->opt.optlen); | |
1070 | ipc.opt = &opt_copy.opt; | |
1071 | } | |
1072 | rcu_read_unlock(); | |
1073 | } | |
db8dac20 | 1074 | |
1cedee13 AI |
1075 | if (cgroup_bpf_enabled && !connected) { |
1076 | err = BPF_CGROUP_RUN_PROG_UDP4_SENDMSG_LOCK(sk, | |
1077 | (struct sockaddr *)usin, &ipc.addr); | |
1078 | if (err) | |
1079 | goto out_free; | |
1080 | if (usin) { | |
1081 | if (usin->sin_port == 0) { | |
1082 | /* BPF program set invalid port. Reject it. */ | |
1083 | err = -EINVAL; | |
1084 | goto out_free; | |
1085 | } | |
1086 | daddr = usin->sin_addr.s_addr; | |
1087 | dport = usin->sin_port; | |
1088 | } | |
1089 | } | |
1090 | ||
db8dac20 DM |
1091 | saddr = ipc.addr; |
1092 | ipc.addr = faddr = daddr; | |
1093 | ||
f6d8bd05 | 1094 | if (ipc.opt && ipc.opt->opt.srr) { |
1b97013b AI |
1095 | if (!daddr) { |
1096 | err = -EINVAL; | |
1097 | goto out_free; | |
1098 | } | |
f6d8bd05 | 1099 | faddr = ipc.opt->opt.faddr; |
db8dac20 DM |
1100 | connected = 0; |
1101 | } | |
aa661581 | 1102 | tos = get_rttos(&ipc, inet); |
db8dac20 DM |
1103 | if (sock_flag(sk, SOCK_LOCALROUTE) || |
1104 | (msg->msg_flags & MSG_DONTROUTE) || | |
f6d8bd05 | 1105 | (ipc.opt && ipc.opt->opt.is_strictroute)) { |
db8dac20 DM |
1106 | tos |= RTO_ONLINK; |
1107 | connected = 0; | |
1108 | } | |
1109 | ||
1110 | if (ipv4_is_multicast(daddr)) { | |
854da991 | 1111 | if (!ipc.oif || netif_index_is_l3_master(sock_net(sk), ipc.oif)) |
db8dac20 DM |
1112 | ipc.oif = inet->mc_index; |
1113 | if (!saddr) | |
1114 | saddr = inet->mc_addr; | |
1115 | connected = 0; | |
9515a2e0 | 1116 | } else if (!ipc.oif) { |
76e21053 | 1117 | ipc.oif = inet->uc_index; |
9515a2e0 DA |
1118 | } else if (ipv4_is_lbcast(daddr) && inet->uc_index) { |
1119 | /* oif is set, packet is to local broadcast and | |
1120 | * and uc_index is set. oif is most likely set | |
1121 | * by sk_bound_dev_if. If uc_index != oif check if the | |
1122 | * oif is an L3 master and uc_index is an L3 slave. | |
1123 | * If so, we want to allow the send using the uc_index. | |
1124 | */ | |
1125 | if (ipc.oif != inet->uc_index && | |
1126 | ipc.oif == l3mdev_master_ifindex_by_index(sock_net(sk), | |
1127 | inet->uc_index)) { | |
1128 | ipc.oif = inet->uc_index; | |
1129 | } | |
1130 | } | |
db8dac20 DM |
1131 | |
1132 | if (connected) | |
c482c568 | 1133 | rt = (struct rtable *)sk_dst_check(sk, 0); |
db8dac20 | 1134 | |
51456b29 | 1135 | if (!rt) { |
84a3aa00 | 1136 | struct net *net = sock_net(sk); |
9a24abfa | 1137 | __u8 flow_flags = inet_sk_flowi_flags(sk); |
84a3aa00 | 1138 | |
e474995f | 1139 | fl4 = &fl4_stack; |
9a24abfa | 1140 | |
e474995f | 1141 | flowi4_init_output(fl4, ipc.oif, sk->sk_mark, tos, |
c0951cbc | 1142 | RT_SCOPE_UNIVERSE, sk->sk_protocol, |
9a24abfa | 1143 | flow_flags, |
e2d118a1 LC |
1144 | faddr, saddr, dport, inet->inet_sport, |
1145 | sk->sk_uid); | |
c0951cbc | 1146 | |
e474995f DM |
1147 | security_sk_classify_flow(sk, flowi4_to_flowi(fl4)); |
1148 | rt = ip_route_output_flow(net, fl4, sk); | |
b23dd4fe DM |
1149 | if (IS_ERR(rt)) { |
1150 | err = PTR_ERR(rt); | |
06dc94b1 | 1151 | rt = NULL; |
db8dac20 | 1152 | if (err == -ENETUNREACH) |
f1d8cba6 | 1153 | IP_INC_STATS(net, IPSTATS_MIB_OUTNOROUTES); |
db8dac20 DM |
1154 | goto out; |
1155 | } | |
1156 | ||
1157 | err = -EACCES; | |
1158 | if ((rt->rt_flags & RTCF_BROADCAST) && | |
1159 | !sock_flag(sk, SOCK_BROADCAST)) | |
1160 | goto out; | |
1161 | if (connected) | |
d8d1f30b | 1162 | sk_dst_set(sk, dst_clone(&rt->dst)); |
db8dac20 DM |
1163 | } |
1164 | ||
1165 | if (msg->msg_flags&MSG_CONFIRM) | |
1166 | goto do_confirm; | |
1167 | back_from_confirm: | |
1168 | ||
e474995f | 1169 | saddr = fl4->saddr; |
db8dac20 | 1170 | if (!ipc.addr) |
e474995f | 1171 | daddr = ipc.addr = fl4->daddr; |
db8dac20 | 1172 | |
903ab86d HX |
1173 | /* Lockless fast path for the non-corking case. */ |
1174 | if (!corkreq) { | |
1cd7884d WB |
1175 | struct inet_cork cork; |
1176 | ||
f69e6d13 | 1177 | skb = ip_make_skb(sk, fl4, getfrag, msg, ulen, |
903ab86d | 1178 | sizeof(struct udphdr), &ipc, &rt, |
1cd7884d | 1179 | &cork, msg->msg_flags); |
903ab86d | 1180 | err = PTR_ERR(skb); |
50c3a487 | 1181 | if (!IS_ERR_OR_NULL(skb)) |
bec1f6f6 | 1182 | err = udp_send_skb(skb, fl4, &cork); |
903ab86d HX |
1183 | goto out; |
1184 | } | |
1185 | ||
db8dac20 DM |
1186 | lock_sock(sk); |
1187 | if (unlikely(up->pending)) { | |
1188 | /* The socket is already corked while preparing it. */ | |
1189 | /* ... which is an evident application bug. --ANK */ | |
1190 | release_sock(sk); | |
1191 | ||
197df02c | 1192 | net_dbg_ratelimited("socket already corked\n"); |
db8dac20 DM |
1193 | err = -EINVAL; |
1194 | goto out; | |
1195 | } | |
1196 | /* | |
1197 | * Now cork the socket to pend data. | |
1198 | */ | |
b6f21b26 DM |
1199 | fl4 = &inet->cork.fl.u.ip4; |
1200 | fl4->daddr = daddr; | |
1201 | fl4->saddr = saddr; | |
9cce96df DM |
1202 | fl4->fl4_dport = dport; |
1203 | fl4->fl4_sport = inet->inet_sport; | |
db8dac20 DM |
1204 | up->pending = AF_INET; |
1205 | ||
1206 | do_append_data: | |
1207 | up->len += ulen; | |
f69e6d13 | 1208 | err = ip_append_data(sk, fl4, getfrag, msg, ulen, |
f5fca608 DM |
1209 | sizeof(struct udphdr), &ipc, &rt, |
1210 | corkreq ? msg->msg_flags|MSG_MORE : msg->msg_flags); | |
db8dac20 DM |
1211 | if (err) |
1212 | udp_flush_pending_frames(sk); | |
1213 | else if (!corkreq) | |
1214 | err = udp_push_pending_frames(sk); | |
1215 | else if (unlikely(skb_queue_empty(&sk->sk_write_queue))) | |
1216 | up->pending = 0; | |
1217 | release_sock(sk); | |
1218 | ||
1219 | out: | |
1220 | ip_rt_put(rt); | |
1b97013b | 1221 | out_free: |
db8dac20 DM |
1222 | if (free) |
1223 | kfree(ipc.opt); | |
1224 | if (!err) | |
1225 | return len; | |
1226 | /* | |
1227 | * ENOBUFS = no kernel mem, SOCK_NOSPACE = no sndbuf space. Reporting | |
1228 | * ENOBUFS might not be good (it's not tunable per se), but otherwise | |
1229 | * we don't have a good statistic (IpOutDiscards but it can be too many | |
1230 | * things). We could add another new stat but at least for now that | |
1231 | * seems like overkill. | |
1232 | */ | |
1233 | if (err == -ENOBUFS || test_bit(SOCK_NOSPACE, &sk->sk_socket->flags)) { | |
6aef70a8 ED |
1234 | UDP_INC_STATS(sock_net(sk), |
1235 | UDP_MIB_SNDBUFERRORS, is_udplite); | |
db8dac20 DM |
1236 | } |
1237 | return err; | |
1238 | ||
1239 | do_confirm: | |
0dec879f JA |
1240 | if (msg->msg_flags & MSG_PROBE) |
1241 | dst_confirm_neigh(&rt->dst, &fl4->daddr); | |
db8dac20 DM |
1242 | if (!(msg->msg_flags&MSG_PROBE) || len) |
1243 | goto back_from_confirm; | |
1244 | err = 0; | |
1245 | goto out; | |
1246 | } | |
c482c568 | 1247 | EXPORT_SYMBOL(udp_sendmsg); |
db8dac20 DM |
1248 | |
1249 | int udp_sendpage(struct sock *sk, struct page *page, int offset, | |
1250 | size_t size, int flags) | |
1251 | { | |
f5fca608 | 1252 | struct inet_sock *inet = inet_sk(sk); |
db8dac20 DM |
1253 | struct udp_sock *up = udp_sk(sk); |
1254 | int ret; | |
1255 | ||
d3f7d56a SL |
1256 | if (flags & MSG_SENDPAGE_NOTLAST) |
1257 | flags |= MSG_MORE; | |
1258 | ||
db8dac20 DM |
1259 | if (!up->pending) { |
1260 | struct msghdr msg = { .msg_flags = flags|MSG_MORE }; | |
1261 | ||
1262 | /* Call udp_sendmsg to specify destination address which | |
1263 | * sendpage interface can't pass. | |
1264 | * This will succeed only when the socket is connected. | |
1265 | */ | |
1b784140 | 1266 | ret = udp_sendmsg(sk, &msg, 0); |
db8dac20 DM |
1267 | if (ret < 0) |
1268 | return ret; | |
1269 | } | |
1270 | ||
1271 | lock_sock(sk); | |
1272 | ||
1273 | if (unlikely(!up->pending)) { | |
1274 | release_sock(sk); | |
1275 | ||
197df02c | 1276 | net_dbg_ratelimited("cork failed\n"); |
db8dac20 DM |
1277 | return -EINVAL; |
1278 | } | |
1279 | ||
f5fca608 DM |
1280 | ret = ip_append_page(sk, &inet->cork.fl.u.ip4, |
1281 | page, offset, size, flags); | |
db8dac20 DM |
1282 | if (ret == -EOPNOTSUPP) { |
1283 | release_sock(sk); | |
1284 | return sock_no_sendpage(sk->sk_socket, page, offset, | |
1285 | size, flags); | |
1286 | } | |
1287 | if (ret < 0) { | |
1288 | udp_flush_pending_frames(sk); | |
1289 | goto out; | |
1290 | } | |
1291 | ||
1292 | up->len += size; | |
1293 | if (!(up->corkflag || (flags&MSG_MORE))) | |
1294 | ret = udp_push_pending_frames(sk); | |
1295 | if (!ret) | |
1296 | ret = size; | |
1297 | out: | |
1298 | release_sock(sk); | |
1299 | return ret; | |
1300 | } | |
1301 | ||
dce4551c PA |
1302 | #define UDP_SKB_IS_STATELESS 0x80000000 |
1303 | ||
b65ac446 PA |
1304 | static void udp_set_dev_scratch(struct sk_buff *skb) |
1305 | { | |
dce4551c | 1306 | struct udp_dev_scratch *scratch = udp_skb_scratch(skb); |
b65ac446 PA |
1307 | |
1308 | BUILD_BUG_ON(sizeof(struct udp_dev_scratch) > sizeof(long)); | |
dce4551c PA |
1309 | scratch->_tsize_state = skb->truesize; |
1310 | #if BITS_PER_LONG == 64 | |
b65ac446 PA |
1311 | scratch->len = skb->len; |
1312 | scratch->csum_unnecessary = !!skb_csum_unnecessary(skb); | |
1313 | scratch->is_linear = !skb_is_nonlinear(skb); | |
dce4551c | 1314 | #endif |
3bdefdf9 PA |
1315 | /* all head states execept sp (dst, sk, nf) are always cleared by |
1316 | * udp_rcv() and we need to preserve secpath, if present, to eventually | |
1317 | * process IP_CMSG_PASSSEC at recvmsg() time | |
1318 | */ | |
1319 | if (likely(!skb_sec_path(skb))) | |
dce4551c | 1320 | scratch->_tsize_state |= UDP_SKB_IS_STATELESS; |
b65ac446 PA |
1321 | } |
1322 | ||
1323 | static int udp_skb_truesize(struct sk_buff *skb) | |
1324 | { | |
dce4551c | 1325 | return udp_skb_scratch(skb)->_tsize_state & ~UDP_SKB_IS_STATELESS; |
b65ac446 PA |
1326 | } |
1327 | ||
dce4551c | 1328 | static bool udp_skb_has_head_state(struct sk_buff *skb) |
b65ac446 | 1329 | { |
dce4551c | 1330 | return !(udp_skb_scratch(skb)->_tsize_state & UDP_SKB_IS_STATELESS); |
b65ac446 | 1331 | } |
b65ac446 | 1332 | |
7c13f97f | 1333 | /* fully reclaim rmem/fwd memory allocated for skb */ |
6dfb4367 PA |
1334 | static void udp_rmem_release(struct sock *sk, int size, int partial, |
1335 | bool rx_queue_lock_held) | |
f970bd9e | 1336 | { |
6b229cf7 | 1337 | struct udp_sock *up = udp_sk(sk); |
2276f58a | 1338 | struct sk_buff_head *sk_queue; |
f970bd9e PA |
1339 | int amt; |
1340 | ||
6b229cf7 ED |
1341 | if (likely(partial)) { |
1342 | up->forward_deficit += size; | |
1343 | size = up->forward_deficit; | |
0d4a6608 | 1344 | if (size < (sk->sk_rcvbuf >> 2)) |
6b229cf7 ED |
1345 | return; |
1346 | } else { | |
1347 | size += up->forward_deficit; | |
1348 | } | |
1349 | up->forward_deficit = 0; | |
1350 | ||
6dfb4367 PA |
1351 | /* acquire the sk_receive_queue for fwd allocated memory scheduling, |
1352 | * if the called don't held it already | |
1353 | */ | |
2276f58a | 1354 | sk_queue = &sk->sk_receive_queue; |
6dfb4367 PA |
1355 | if (!rx_queue_lock_held) |
1356 | spin_lock(&sk_queue->lock); | |
1357 | ||
2276f58a | 1358 | |
f970bd9e PA |
1359 | sk->sk_forward_alloc += size; |
1360 | amt = (sk->sk_forward_alloc - partial) & ~(SK_MEM_QUANTUM - 1); | |
1361 | sk->sk_forward_alloc -= amt; | |
f970bd9e PA |
1362 | |
1363 | if (amt) | |
1364 | __sk_mem_reduce_allocated(sk, amt >> SK_MEM_QUANTUM_SHIFT); | |
02ab0d13 ED |
1365 | |
1366 | atomic_sub(size, &sk->sk_rmem_alloc); | |
2276f58a PA |
1367 | |
1368 | /* this can save us from acquiring the rx queue lock on next receive */ | |
1369 | skb_queue_splice_tail_init(sk_queue, &up->reader_queue); | |
1370 | ||
6dfb4367 PA |
1371 | if (!rx_queue_lock_held) |
1372 | spin_unlock(&sk_queue->lock); | |
f970bd9e PA |
1373 | } |
1374 | ||
2276f58a | 1375 | /* Note: called with reader_queue.lock held. |
c84d9490 ED |
1376 | * Instead of using skb->truesize here, find a copy of it in skb->dev_scratch |
1377 | * This avoids a cache line miss while receive_queue lock is held. | |
1378 | * Look at __udp_enqueue_schedule_skb() to find where this copy is done. | |
1379 | */ | |
7c13f97f | 1380 | void udp_skb_destructor(struct sock *sk, struct sk_buff *skb) |
f970bd9e | 1381 | { |
b65ac446 PA |
1382 | prefetch(&skb->data); |
1383 | udp_rmem_release(sk, udp_skb_truesize(skb), 1, false); | |
f970bd9e | 1384 | } |
7c13f97f | 1385 | EXPORT_SYMBOL(udp_skb_destructor); |
f970bd9e | 1386 | |
6dfb4367 | 1387 | /* as above, but the caller held the rx queue lock, too */ |
64f5102d | 1388 | static void udp_skb_dtor_locked(struct sock *sk, struct sk_buff *skb) |
6dfb4367 | 1389 | { |
b65ac446 PA |
1390 | prefetch(&skb->data); |
1391 | udp_rmem_release(sk, udp_skb_truesize(skb), 1, true); | |
6dfb4367 PA |
1392 | } |
1393 | ||
4b272750 ED |
1394 | /* Idea of busylocks is to let producers grab an extra spinlock |
1395 | * to relieve pressure on the receive_queue spinlock shared by consumer. | |
1396 | * Under flood, this means that only one producer can be in line | |
1397 | * trying to acquire the receive_queue spinlock. | |
1398 | * These busylock can be allocated on a per cpu manner, instead of a | |
1399 | * per socket one (that would consume a cache line per socket) | |
1400 | */ | |
1401 | static int udp_busylocks_log __read_mostly; | |
1402 | static spinlock_t *udp_busylocks __read_mostly; | |
1403 | ||
1404 | static spinlock_t *busylock_acquire(void *ptr) | |
1405 | { | |
1406 | spinlock_t *busy; | |
1407 | ||
1408 | busy = udp_busylocks + hash_ptr(ptr, udp_busylocks_log); | |
1409 | spin_lock(busy); | |
1410 | return busy; | |
1411 | } | |
1412 | ||
1413 | static void busylock_release(spinlock_t *busy) | |
1414 | { | |
1415 | if (busy) | |
1416 | spin_unlock(busy); | |
1417 | } | |
1418 | ||
f970bd9e PA |
1419 | int __udp_enqueue_schedule_skb(struct sock *sk, struct sk_buff *skb) |
1420 | { | |
1421 | struct sk_buff_head *list = &sk->sk_receive_queue; | |
1422 | int rmem, delta, amt, err = -ENOMEM; | |
4b272750 | 1423 | spinlock_t *busy = NULL; |
c8c8b127 | 1424 | int size; |
f970bd9e PA |
1425 | |
1426 | /* try to avoid the costly atomic add/sub pair when the receive | |
1427 | * queue is full; always allow at least a packet | |
1428 | */ | |
1429 | rmem = atomic_read(&sk->sk_rmem_alloc); | |
363dc73a | 1430 | if (rmem > sk->sk_rcvbuf) |
f970bd9e PA |
1431 | goto drop; |
1432 | ||
c8c8b127 ED |
1433 | /* Under mem pressure, it might be helpful to help udp_recvmsg() |
1434 | * having linear skbs : | |
1435 | * - Reduce memory overhead and thus increase receive queue capacity | |
1436 | * - Less cache line misses at copyout() time | |
1437 | * - Less work at consume_skb() (less alien page frag freeing) | |
1438 | */ | |
4b272750 | 1439 | if (rmem > (sk->sk_rcvbuf >> 1)) { |
c8c8b127 | 1440 | skb_condense(skb); |
4b272750 ED |
1441 | |
1442 | busy = busylock_acquire(sk); | |
1443 | } | |
c8c8b127 | 1444 | size = skb->truesize; |
b65ac446 | 1445 | udp_set_dev_scratch(skb); |
c8c8b127 | 1446 | |
f970bd9e PA |
1447 | /* we drop only if the receive buf is full and the receive |
1448 | * queue contains some other skb | |
1449 | */ | |
1450 | rmem = atomic_add_return(size, &sk->sk_rmem_alloc); | |
363dc73a | 1451 | if (rmem > (size + sk->sk_rcvbuf)) |
f970bd9e PA |
1452 | goto uncharge_drop; |
1453 | ||
1454 | spin_lock(&list->lock); | |
1455 | if (size >= sk->sk_forward_alloc) { | |
1456 | amt = sk_mem_pages(size); | |
1457 | delta = amt << SK_MEM_QUANTUM_SHIFT; | |
1458 | if (!__sk_mem_raise_allocated(sk, delta, amt, SK_MEM_RECV)) { | |
1459 | err = -ENOBUFS; | |
1460 | spin_unlock(&list->lock); | |
1461 | goto uncharge_drop; | |
1462 | } | |
1463 | ||
1464 | sk->sk_forward_alloc += delta; | |
1465 | } | |
1466 | ||
1467 | sk->sk_forward_alloc -= size; | |
1468 | ||
7c13f97f PA |
1469 | /* no need to setup a destructor, we will explicitly release the |
1470 | * forward allocated memory on dequeue | |
1471 | */ | |
f970bd9e PA |
1472 | sock_skb_set_dropcount(sk, skb); |
1473 | ||
1474 | __skb_queue_tail(list, skb); | |
1475 | spin_unlock(&list->lock); | |
1476 | ||
1477 | if (!sock_flag(sk, SOCK_DEAD)) | |
1478 | sk->sk_data_ready(sk); | |
1479 | ||
4b272750 | 1480 | busylock_release(busy); |
f970bd9e PA |
1481 | return 0; |
1482 | ||
1483 | uncharge_drop: | |
1484 | atomic_sub(skb->truesize, &sk->sk_rmem_alloc); | |
1485 | ||
1486 | drop: | |
1487 | atomic_inc(&sk->sk_drops); | |
4b272750 | 1488 | busylock_release(busy); |
f970bd9e PA |
1489 | return err; |
1490 | } | |
1491 | EXPORT_SYMBOL_GPL(__udp_enqueue_schedule_skb); | |
1492 | ||
c915fe13 | 1493 | void udp_destruct_sock(struct sock *sk) |
f970bd9e PA |
1494 | { |
1495 | /* reclaim completely the forward allocated memory */ | |
2276f58a | 1496 | struct udp_sock *up = udp_sk(sk); |
7c13f97f PA |
1497 | unsigned int total = 0; |
1498 | struct sk_buff *skb; | |
1499 | ||
2276f58a PA |
1500 | skb_queue_splice_tail_init(&sk->sk_receive_queue, &up->reader_queue); |
1501 | while ((skb = __skb_dequeue(&up->reader_queue)) != NULL) { | |
7c13f97f PA |
1502 | total += skb->truesize; |
1503 | kfree_skb(skb); | |
1504 | } | |
6dfb4367 | 1505 | udp_rmem_release(sk, total, 0, true); |
7c13f97f | 1506 | |
f970bd9e PA |
1507 | inet_sock_destruct(sk); |
1508 | } | |
c915fe13 | 1509 | EXPORT_SYMBOL_GPL(udp_destruct_sock); |
f970bd9e PA |
1510 | |
1511 | int udp_init_sock(struct sock *sk) | |
1512 | { | |
2276f58a | 1513 | skb_queue_head_init(&udp_sk(sk)->reader_queue); |
f970bd9e PA |
1514 | sk->sk_destruct = udp_destruct_sock; |
1515 | return 0; | |
1516 | } | |
1517 | EXPORT_SYMBOL_GPL(udp_init_sock); | |
1518 | ||
1519 | void skb_consume_udp(struct sock *sk, struct sk_buff *skb, int len) | |
1520 | { | |
1521 | if (unlikely(READ_ONCE(sk->sk_peek_off) >= 0)) { | |
1522 | bool slow = lock_sock_fast(sk); | |
1523 | ||
1524 | sk_peek_offset_bwd(sk, len); | |
1525 | unlock_sock_fast(sk, slow); | |
1526 | } | |
0a463c78 | 1527 | |
ca2c1418 PA |
1528 | if (!skb_unref(skb)) |
1529 | return; | |
1530 | ||
dce4551c PA |
1531 | /* In the more common cases we cleared the head states previously, |
1532 | * see __udp_queue_rcv_skb(). | |
0ddf3fb2 | 1533 | */ |
dce4551c | 1534 | if (unlikely(udp_skb_has_head_state(skb))) |
0ddf3fb2 | 1535 | skb_release_head_state(skb); |
ca2c1418 | 1536 | __consume_stateless_skb(skb); |
f970bd9e PA |
1537 | } |
1538 | EXPORT_SYMBOL_GPL(skb_consume_udp); | |
1539 | ||
2276f58a PA |
1540 | static struct sk_buff *__first_packet_length(struct sock *sk, |
1541 | struct sk_buff_head *rcvq, | |
1542 | int *total) | |
1543 | { | |
1544 | struct sk_buff *skb; | |
1545 | ||
9bd780f5 PA |
1546 | while ((skb = skb_peek(rcvq)) != NULL) { |
1547 | if (udp_lib_checksum_complete(skb)) { | |
1548 | __UDP_INC_STATS(sock_net(sk), UDP_MIB_CSUMERRORS, | |
1549 | IS_UDPLITE(sk)); | |
1550 | __UDP_INC_STATS(sock_net(sk), UDP_MIB_INERRORS, | |
1551 | IS_UDPLITE(sk)); | |
1552 | atomic_inc(&sk->sk_drops); | |
1553 | __skb_unlink(skb, rcvq); | |
1554 | *total += skb->truesize; | |
1555 | kfree_skb(skb); | |
1556 | } else { | |
1557 | /* the csum related bits could be changed, refresh | |
1558 | * the scratch area | |
1559 | */ | |
1560 | udp_set_dev_scratch(skb); | |
1561 | break; | |
1562 | } | |
2276f58a PA |
1563 | } |
1564 | return skb; | |
1565 | } | |
1566 | ||
85584672 ED |
1567 | /** |
1568 | * first_packet_length - return length of first packet in receive queue | |
1569 | * @sk: socket | |
1570 | * | |
1571 | * Drops all bad checksum frames, until a valid one is found. | |
e83c6744 | 1572 | * Returns the length of found skb, or -1 if none is found. |
85584672 | 1573 | */ |
e83c6744 | 1574 | static int first_packet_length(struct sock *sk) |
85584672 | 1575 | { |
2276f58a PA |
1576 | struct sk_buff_head *rcvq = &udp_sk(sk)->reader_queue; |
1577 | struct sk_buff_head *sk_queue = &sk->sk_receive_queue; | |
85584672 | 1578 | struct sk_buff *skb; |
7c13f97f | 1579 | int total = 0; |
e83c6744 | 1580 | int res; |
85584672 | 1581 | |
85584672 | 1582 | spin_lock_bh(&rcvq->lock); |
2276f58a PA |
1583 | skb = __first_packet_length(sk, rcvq, &total); |
1584 | if (!skb && !skb_queue_empty(sk_queue)) { | |
1585 | spin_lock(&sk_queue->lock); | |
1586 | skb_queue_splice_tail_init(sk_queue, rcvq); | |
1587 | spin_unlock(&sk_queue->lock); | |
1588 | ||
1589 | skb = __first_packet_length(sk, rcvq, &total); | |
85584672 | 1590 | } |
e83c6744 | 1591 | res = skb ? skb->len : -1; |
7c13f97f | 1592 | if (total) |
6dfb4367 | 1593 | udp_rmem_release(sk, total, 1, false); |
85584672 | 1594 | spin_unlock_bh(&rcvq->lock); |
85584672 ED |
1595 | return res; |
1596 | } | |
1597 | ||
1da177e4 LT |
1598 | /* |
1599 | * IOCTL requests applicable to the UDP protocol | |
1600 | */ | |
e905a9ed | 1601 | |
1da177e4 LT |
1602 | int udp_ioctl(struct sock *sk, int cmd, unsigned long arg) |
1603 | { | |
6516c655 SH |
1604 | switch (cmd) { |
1605 | case SIOCOUTQ: | |
1da177e4 | 1606 | { |
31e6d363 ED |
1607 | int amount = sk_wmem_alloc_get(sk); |
1608 | ||
6516c655 SH |
1609 | return put_user(amount, (int __user *)arg); |
1610 | } | |
1da177e4 | 1611 | |
6516c655 SH |
1612 | case SIOCINQ: |
1613 | { | |
e83c6744 | 1614 | int amount = max_t(int, 0, first_packet_length(sk)); |
6516c655 | 1615 | |
6516c655 SH |
1616 | return put_user(amount, (int __user *)arg); |
1617 | } | |
1da177e4 | 1618 | |
6516c655 SH |
1619 | default: |
1620 | return -ENOIOCTLCMD; | |
1da177e4 | 1621 | } |
6516c655 SH |
1622 | |
1623 | return 0; | |
1da177e4 | 1624 | } |
c482c568 | 1625 | EXPORT_SYMBOL(udp_ioctl); |
1da177e4 | 1626 | |
2276f58a | 1627 | struct sk_buff *__skb_recv_udp(struct sock *sk, unsigned int flags, |
fd69c399 | 1628 | int noblock, int *off, int *err) |
2276f58a PA |
1629 | { |
1630 | struct sk_buff_head *sk_queue = &sk->sk_receive_queue; | |
1631 | struct sk_buff_head *queue; | |
1632 | struct sk_buff *last; | |
1633 | long timeo; | |
1634 | int error; | |
1635 | ||
1636 | queue = &udp_sk(sk)->reader_queue; | |
1637 | flags |= noblock ? MSG_DONTWAIT : 0; | |
1638 | timeo = sock_rcvtimeo(sk, flags & MSG_DONTWAIT); | |
1639 | do { | |
1640 | struct sk_buff *skb; | |
1641 | ||
1642 | error = sock_error(sk); | |
1643 | if (error) | |
1644 | break; | |
1645 | ||
1646 | error = -EAGAIN; | |
2276f58a | 1647 | do { |
2276f58a PA |
1648 | spin_lock_bh(&queue->lock); |
1649 | skb = __skb_try_recv_from_queue(sk, queue, flags, | |
1650 | udp_skb_destructor, | |
fd69c399 | 1651 | off, err, &last); |
2276f58a PA |
1652 | if (skb) { |
1653 | spin_unlock_bh(&queue->lock); | |
2276f58a PA |
1654 | return skb; |
1655 | } | |
1656 | ||
1657 | if (skb_queue_empty(sk_queue)) { | |
1658 | spin_unlock_bh(&queue->lock); | |
1659 | goto busy_check; | |
1660 | } | |
1661 | ||
6dfb4367 PA |
1662 | /* refill the reader queue and walk it again |
1663 | * keep both queues locked to avoid re-acquiring | |
1664 | * the sk_receive_queue lock if fwd memory scheduling | |
1665 | * is needed. | |
1666 | */ | |
2276f58a PA |
1667 | spin_lock(&sk_queue->lock); |
1668 | skb_queue_splice_tail_init(sk_queue, queue); | |
2276f58a PA |
1669 | |
1670 | skb = __skb_try_recv_from_queue(sk, queue, flags, | |
6dfb4367 | 1671 | udp_skb_dtor_locked, |
fd69c399 | 1672 | off, err, &last); |
6dfb4367 | 1673 | spin_unlock(&sk_queue->lock); |
2276f58a | 1674 | spin_unlock_bh(&queue->lock); |
de321ed3 | 1675 | if (skb) |
2276f58a | 1676 | return skb; |
2276f58a PA |
1677 | |
1678 | busy_check: | |
1679 | if (!sk_can_busy_loop(sk)) | |
1680 | break; | |
1681 | ||
1682 | sk_busy_loop(sk, flags & MSG_DONTWAIT); | |
1683 | } while (!skb_queue_empty(sk_queue)); | |
1684 | ||
1685 | /* sk_queue is empty, reader_queue may contain peeked packets */ | |
1686 | } while (timeo && | |
1687 | !__skb_wait_for_more_packets(sk, &error, &timeo, | |
1688 | (struct sk_buff *)sk_queue)); | |
1689 | ||
1690 | *err = error; | |
1691 | return NULL; | |
1692 | } | |
7e823644 | 1693 | EXPORT_SYMBOL(__skb_recv_udp); |
2276f58a | 1694 | |
db8dac20 DM |
1695 | /* |
1696 | * This should be easy, if there is something there we | |
1697 | * return it, otherwise we block. | |
1698 | */ | |
1699 | ||
1b784140 YX |
1700 | int udp_recvmsg(struct sock *sk, struct msghdr *msg, size_t len, int noblock, |
1701 | int flags, int *addr_len) | |
db8dac20 DM |
1702 | { |
1703 | struct inet_sock *inet = inet_sk(sk); | |
342dfc30 | 1704 | DECLARE_SOCKADDR(struct sockaddr_in *, sin, msg->msg_name); |
db8dac20 | 1705 | struct sk_buff *skb; |
59c2cdae | 1706 | unsigned int ulen, copied; |
fd69c399 | 1707 | int off, err, peeking = flags & MSG_PEEK; |
db8dac20 | 1708 | int is_udplite = IS_UDPLITE(sk); |
197c949e | 1709 | bool checksum_valid = false; |
db8dac20 | 1710 | |
db8dac20 | 1711 | if (flags & MSG_ERRQUEUE) |
85fbaa75 | 1712 | return ip_recv_error(sk, msg, len, addr_len); |
db8dac20 DM |
1713 | |
1714 | try_again: | |
a0917e0b | 1715 | off = sk_peek_offset(sk, flags); |
fd69c399 | 1716 | skb = __skb_recv_udp(sk, flags, noblock, &off, &err); |
db8dac20 | 1717 | if (!skb) |
627d2d6b | 1718 | return err; |
db8dac20 | 1719 | |
b65ac446 | 1720 | ulen = udp_skb_len(skb); |
59c2cdae | 1721 | copied = len; |
627d2d6b | 1722 | if (copied > ulen - off) |
1723 | copied = ulen - off; | |
59c2cdae | 1724 | else if (copied < ulen) |
db8dac20 DM |
1725 | msg->msg_flags |= MSG_TRUNC; |
1726 | ||
1727 | /* | |
1728 | * If checksum is needed at all, try to do it while copying the | |
1729 | * data. If the data is truncated, or if we only want a partial | |
1730 | * coverage checksum (UDP-Lite), do it before the copy. | |
1731 | */ | |
1732 | ||
d21dbdfe ED |
1733 | if (copied < ulen || peeking || |
1734 | (is_udplite && UDP_SKB_CB(skb)->partial_cov)) { | |
b65ac446 PA |
1735 | checksum_valid = udp_skb_csum_unnecessary(skb) || |
1736 | !__udp_lib_checksum_complete(skb); | |
197c949e | 1737 | if (!checksum_valid) |
db8dac20 DM |
1738 | goto csum_copy_err; |
1739 | } | |
1740 | ||
b65ac446 PA |
1741 | if (checksum_valid || udp_skb_csum_unnecessary(skb)) { |
1742 | if (udp_skb_is_linear(skb)) | |
1743 | err = copy_linear_skb(skb, copied, off, &msg->msg_iter); | |
1744 | else | |
1745 | err = skb_copy_datagram_msg(skb, off, msg, copied); | |
1746 | } else { | |
627d2d6b | 1747 | err = skb_copy_and_csum_datagram_msg(skb, off, msg); |
db8dac20 DM |
1748 | |
1749 | if (err == -EINVAL) | |
1750 | goto csum_copy_err; | |
1751 | } | |
1752 | ||
22911fc5 | 1753 | if (unlikely(err)) { |
fd69c399 | 1754 | if (!peeking) { |
979402b1 | 1755 | atomic_inc(&sk->sk_drops); |
6aef70a8 ED |
1756 | UDP_INC_STATS(sock_net(sk), |
1757 | UDP_MIB_INERRORS, is_udplite); | |
979402b1 | 1758 | } |
850cbadd | 1759 | kfree_skb(skb); |
627d2d6b | 1760 | return err; |
22911fc5 | 1761 | } |
db8dac20 | 1762 | |
fd69c399 | 1763 | if (!peeking) |
6aef70a8 ED |
1764 | UDP_INC_STATS(sock_net(sk), |
1765 | UDP_MIB_INDATAGRAMS, is_udplite); | |
db8dac20 | 1766 | |
3b885787 | 1767 | sock_recv_ts_and_drops(msg, sk, skb); |
db8dac20 DM |
1768 | |
1769 | /* Copy the address. */ | |
c482c568 | 1770 | if (sin) { |
db8dac20 DM |
1771 | sin->sin_family = AF_INET; |
1772 | sin->sin_port = udp_hdr(skb)->source; | |
1773 | sin->sin_addr.s_addr = ip_hdr(skb)->saddr; | |
1774 | memset(sin->sin_zero, 0, sizeof(sin->sin_zero)); | |
bceaa902 | 1775 | *addr_len = sizeof(*sin); |
db8dac20 | 1776 | } |
bcd1665e PA |
1777 | |
1778 | if (udp_sk(sk)->gro_enabled) | |
1779 | udp_cmsg_recv(msg, sk, skb); | |
1780 | ||
db8dac20 | 1781 | if (inet->cmsg_flags) |
ad959036 | 1782 | ip_cmsg_recv_offset(msg, sk, skb, sizeof(struct udphdr), off); |
db8dac20 | 1783 | |
59c2cdae | 1784 | err = copied; |
db8dac20 DM |
1785 | if (flags & MSG_TRUNC) |
1786 | err = ulen; | |
1787 | ||
850cbadd | 1788 | skb_consume_udp(sk, skb, peeking ? -err : err); |
db8dac20 DM |
1789 | return err; |
1790 | ||
1791 | csum_copy_err: | |
2276f58a PA |
1792 | if (!__sk_queue_drop_skb(sk, &udp_sk(sk)->reader_queue, skb, flags, |
1793 | udp_skb_destructor)) { | |
6aef70a8 ED |
1794 | UDP_INC_STATS(sock_net(sk), UDP_MIB_CSUMERRORS, is_udplite); |
1795 | UDP_INC_STATS(sock_net(sk), UDP_MIB_INERRORS, is_udplite); | |
6a5dc9e5 | 1796 | } |
850cbadd | 1797 | kfree_skb(skb); |
db8dac20 | 1798 | |
beb39db5 ED |
1799 | /* starting over for a new packet, but check if we need to yield */ |
1800 | cond_resched(); | |
9cfaa8de | 1801 | msg->msg_flags &= ~MSG_TRUNC; |
db8dac20 DM |
1802 | goto try_again; |
1803 | } | |
1804 | ||
d74bad4e AI |
1805 | int udp_pre_connect(struct sock *sk, struct sockaddr *uaddr, int addr_len) |
1806 | { | |
1807 | /* This check is replicated from __ip4_datagram_connect() and | |
1808 | * intended to prevent BPF program called below from accessing bytes | |
1809 | * that are out of the bound specified by user in addr_len. | |
1810 | */ | |
1811 | if (addr_len < sizeof(struct sockaddr_in)) | |
1812 | return -EINVAL; | |
1813 | ||
1814 | return BPF_CGROUP_RUN_PROG_INET4_CONNECT_LOCK(sk, uaddr); | |
1815 | } | |
1816 | EXPORT_SYMBOL(udp_pre_connect); | |
1817 | ||
286c72de | 1818 | int __udp_disconnect(struct sock *sk, int flags) |
1da177e4 LT |
1819 | { |
1820 | struct inet_sock *inet = inet_sk(sk); | |
1821 | /* | |
1822 | * 1003.1g - break association. | |
1823 | */ | |
e905a9ed | 1824 | |
1da177e4 | 1825 | sk->sk_state = TCP_CLOSE; |
c720c7e8 ED |
1826 | inet->inet_daddr = 0; |
1827 | inet->inet_dport = 0; | |
bdeab991 | 1828 | sock_rps_reset_rxhash(sk); |
1da177e4 LT |
1829 | sk->sk_bound_dev_if = 0; |
1830 | if (!(sk->sk_userlocks & SOCK_BINDADDR_LOCK)) | |
1831 | inet_reset_saddr(sk); | |
1832 | ||
1833 | if (!(sk->sk_userlocks & SOCK_BINDPORT_LOCK)) { | |
1834 | sk->sk_prot->unhash(sk); | |
c720c7e8 | 1835 | inet->inet_sport = 0; |
1da177e4 LT |
1836 | } |
1837 | sk_dst_reset(sk); | |
1838 | return 0; | |
1839 | } | |
286c72de ED |
1840 | EXPORT_SYMBOL(__udp_disconnect); |
1841 | ||
1842 | int udp_disconnect(struct sock *sk, int flags) | |
1843 | { | |
1844 | lock_sock(sk); | |
1845 | __udp_disconnect(sk, flags); | |
1846 | release_sock(sk); | |
1847 | return 0; | |
1848 | } | |
c482c568 | 1849 | EXPORT_SYMBOL(udp_disconnect); |
1da177e4 | 1850 | |
645ca708 ED |
1851 | void udp_lib_unhash(struct sock *sk) |
1852 | { | |
723b4610 ED |
1853 | if (sk_hashed(sk)) { |
1854 | struct udp_table *udptable = sk->sk_prot->h.udp_table; | |
512615b6 ED |
1855 | struct udp_hslot *hslot, *hslot2; |
1856 | ||
1857 | hslot = udp_hashslot(udptable, sock_net(sk), | |
1858 | udp_sk(sk)->udp_port_hash); | |
1859 | hslot2 = udp_hashslot2(udptable, udp_sk(sk)->udp_portaddr_hash); | |
645ca708 | 1860 | |
723b4610 | 1861 | spin_lock_bh(&hslot->lock); |
e32ea7e7 CG |
1862 | if (rcu_access_pointer(sk->sk_reuseport_cb)) |
1863 | reuseport_detach_sock(sk); | |
ca065d0c | 1864 | if (sk_del_node_init_rcu(sk)) { |
fdcc8aa9 | 1865 | hslot->count--; |
c720c7e8 | 1866 | inet_sk(sk)->inet_num = 0; |
723b4610 | 1867 | sock_prot_inuse_add(sock_net(sk), sk->sk_prot, -1); |
512615b6 ED |
1868 | |
1869 | spin_lock(&hslot2->lock); | |
ca065d0c | 1870 | hlist_del_init_rcu(&udp_sk(sk)->udp_portaddr_node); |
512615b6 ED |
1871 | hslot2->count--; |
1872 | spin_unlock(&hslot2->lock); | |
723b4610 ED |
1873 | } |
1874 | spin_unlock_bh(&hslot->lock); | |
645ca708 | 1875 | } |
645ca708 ED |
1876 | } |
1877 | EXPORT_SYMBOL(udp_lib_unhash); | |
1878 | ||
719f8358 ED |
1879 | /* |
1880 | * inet_rcv_saddr was changed, we must rehash secondary hash | |
1881 | */ | |
1882 | void udp_lib_rehash(struct sock *sk, u16 newhash) | |
1883 | { | |
1884 | if (sk_hashed(sk)) { | |
1885 | struct udp_table *udptable = sk->sk_prot->h.udp_table; | |
1886 | struct udp_hslot *hslot, *hslot2, *nhslot2; | |
1887 | ||
1888 | hslot2 = udp_hashslot2(udptable, udp_sk(sk)->udp_portaddr_hash); | |
1889 | nhslot2 = udp_hashslot2(udptable, newhash); | |
1890 | udp_sk(sk)->udp_portaddr_hash = newhash; | |
e32ea7e7 CG |
1891 | |
1892 | if (hslot2 != nhslot2 || | |
1893 | rcu_access_pointer(sk->sk_reuseport_cb)) { | |
719f8358 ED |
1894 | hslot = udp_hashslot(udptable, sock_net(sk), |
1895 | udp_sk(sk)->udp_port_hash); | |
1896 | /* we must lock primary chain too */ | |
1897 | spin_lock_bh(&hslot->lock); | |
e32ea7e7 CG |
1898 | if (rcu_access_pointer(sk->sk_reuseport_cb)) |
1899 | reuseport_detach_sock(sk); | |
1900 | ||
1901 | if (hslot2 != nhslot2) { | |
1902 | spin_lock(&hslot2->lock); | |
ca065d0c | 1903 | hlist_del_init_rcu(&udp_sk(sk)->udp_portaddr_node); |
e32ea7e7 CG |
1904 | hslot2->count--; |
1905 | spin_unlock(&hslot2->lock); | |
1906 | ||
1907 | spin_lock(&nhslot2->lock); | |
ca065d0c | 1908 | hlist_add_head_rcu(&udp_sk(sk)->udp_portaddr_node, |
e32ea7e7 CG |
1909 | &nhslot2->head); |
1910 | nhslot2->count++; | |
1911 | spin_unlock(&nhslot2->lock); | |
1912 | } | |
719f8358 ED |
1913 | |
1914 | spin_unlock_bh(&hslot->lock); | |
1915 | } | |
1916 | } | |
1917 | } | |
1918 | EXPORT_SYMBOL(udp_lib_rehash); | |
1919 | ||
8f6b5392 | 1920 | void udp_v4_rehash(struct sock *sk) |
719f8358 | 1921 | { |
f0b1e64c | 1922 | u16 new_hash = ipv4_portaddr_hash(sock_net(sk), |
719f8358 ED |
1923 | inet_sk(sk)->inet_rcv_saddr, |
1924 | inet_sk(sk)->inet_num); | |
1925 | udp_lib_rehash(sk, new_hash); | |
1926 | } | |
1927 | ||
a3f96c47 | 1928 | static int __udp_queue_rcv_skb(struct sock *sk, struct sk_buff *skb) |
93821778 | 1929 | { |
fec5e652 | 1930 | int rc; |
766e9037 | 1931 | |
005ec974 | 1932 | if (inet_sk(sk)->inet_daddr) { |
bdeab991 | 1933 | sock_rps_save_rxhash(sk, skb); |
005ec974 | 1934 | sk_mark_napi_id(sk, skb); |
2c8c56e1 | 1935 | sk_incoming_cpu_update(sk); |
e68b6e50 ED |
1936 | } else { |
1937 | sk_mark_napi_id_once(sk, skb); | |
005ec974 | 1938 | } |
fec5e652 | 1939 | |
850cbadd | 1940 | rc = __udp_enqueue_schedule_skb(sk, skb); |
766e9037 ED |
1941 | if (rc < 0) { |
1942 | int is_udplite = IS_UDPLITE(sk); | |
93821778 | 1943 | |
93821778 | 1944 | /* Note that an ENOMEM error is charged twice */ |
766e9037 | 1945 | if (rc == -ENOMEM) |
e61da9e2 | 1946 | UDP_INC_STATS(sock_net(sk), UDP_MIB_RCVBUFERRORS, |
02c22347 | 1947 | is_udplite); |
e61da9e2 | 1948 | UDP_INC_STATS(sock_net(sk), UDP_MIB_INERRORS, is_udplite); |
766e9037 | 1949 | kfree_skb(skb); |
296f7ea7 | 1950 | trace_udp_fail_queue_rcv_skb(rc, sk); |
766e9037 | 1951 | return -1; |
93821778 HX |
1952 | } |
1953 | ||
1954 | return 0; | |
93821778 HX |
1955 | } |
1956 | ||
db8dac20 DM |
1957 | /* returns: |
1958 | * -1: error | |
1959 | * 0: success | |
1960 | * >0: "udp encap" protocol resubmission | |
1961 | * | |
1962 | * Note that in the success and error cases, the skb is assumed to | |
1963 | * have either been requeued or freed. | |
1964 | */ | |
cf329aa4 | 1965 | static int udp_queue_rcv_one_skb(struct sock *sk, struct sk_buff *skb) |
db8dac20 DM |
1966 | { |
1967 | struct udp_sock *up = udp_sk(sk); | |
db8dac20 DM |
1968 | int is_udplite = IS_UDPLITE(sk); |
1969 | ||
1970 | /* | |
1971 | * Charge it to the socket, dropping if the queue is full. | |
1972 | */ | |
1973 | if (!xfrm4_policy_check(sk, XFRM_POLICY_IN, skb)) | |
1974 | goto drop; | |
1975 | nf_reset(skb); | |
1976 | ||
88ab3108 | 1977 | if (static_branch_unlikely(&udp_encap_needed_key) && up->encap_type) { |
0ad92ad0 ED |
1978 | int (*encap_rcv)(struct sock *sk, struct sk_buff *skb); |
1979 | ||
db8dac20 DM |
1980 | /* |
1981 | * This is an encapsulation socket so pass the skb to | |
1982 | * the socket's udp_encap_rcv() hook. Otherwise, just | |
1983 | * fall through and pass this up the UDP socket. | |
1984 | * up->encap_rcv() returns the following value: | |
1985 | * =0 if skb was successfully passed to the encap | |
1986 | * handler or was discarded by it. | |
1987 | * >0 if skb should be passed on to UDP. | |
1988 | * <0 if skb should be resubmitted as proto -N | |
1989 | */ | |
1990 | ||
1991 | /* if we're overly short, let UDP handle it */ | |
6aa7de05 | 1992 | encap_rcv = READ_ONCE(up->encap_rcv); |
e5aed006 | 1993 | if (encap_rcv) { |
db8dac20 DM |
1994 | int ret; |
1995 | ||
0a80966b TH |
1996 | /* Verify checksum before giving to encap */ |
1997 | if (udp_lib_checksum_complete(skb)) | |
1998 | goto csum_error; | |
1999 | ||
0ad92ad0 | 2000 | ret = encap_rcv(sk, skb); |
db8dac20 | 2001 | if (ret <= 0) { |
02c22347 ED |
2002 | __UDP_INC_STATS(sock_net(sk), |
2003 | UDP_MIB_INDATAGRAMS, | |
2004 | is_udplite); | |
db8dac20 DM |
2005 | return -ret; |
2006 | } | |
2007 | } | |
2008 | ||
2009 | /* FALLTHROUGH -- it's a UDP Packet */ | |
2010 | } | |
2011 | ||
2012 | /* | |
2013 | * UDP-Lite specific tests, ignored on UDP sockets | |
2014 | */ | |
2015 | if ((is_udplite & UDPLITE_RECV_CC) && UDP_SKB_CB(skb)->partial_cov) { | |
2016 | ||
2017 | /* | |
2018 | * MIB statistics other than incrementing the error count are | |
2019 | * disabled for the following two types of errors: these depend | |
2020 | * on the application settings, not on the functioning of the | |
2021 | * protocol stack as such. | |
2022 | * | |
2023 | * RFC 3828 here recommends (sec 3.3): "There should also be a | |
2024 | * way ... to ... at least let the receiving application block | |
2025 | * delivery of packets with coverage values less than a value | |
2026 | * provided by the application." | |
2027 | */ | |
2028 | if (up->pcrlen == 0) { /* full coverage was set */ | |
ba7a46f1 JP |
2029 | net_dbg_ratelimited("UDPLite: partial coverage %d while full coverage %d requested\n", |
2030 | UDP_SKB_CB(skb)->cscov, skb->len); | |
db8dac20 DM |
2031 | goto drop; |
2032 | } | |
2033 | /* The next case involves violating the min. coverage requested | |
2034 | * by the receiver. This is subtle: if receiver wants x and x is | |
2035 | * greater than the buffersize/MTU then receiver will complain | |
2036 | * that it wants x while sender emits packets of smaller size y. | |
2037 | * Therefore the above ...()->partial_cov statement is essential. | |
2038 | */ | |
2039 | if (UDP_SKB_CB(skb)->cscov < up->pcrlen) { | |
ba7a46f1 JP |
2040 | net_dbg_ratelimited("UDPLite: coverage %d too small, need min %d\n", |
2041 | UDP_SKB_CB(skb)->cscov, up->pcrlen); | |
db8dac20 DM |
2042 | goto drop; |
2043 | } | |
2044 | } | |
2045 | ||
dd99e425 | 2046 | prefetch(&sk->sk_rmem_alloc); |
ce25d66a ED |
2047 | if (rcu_access_pointer(sk->sk_filter) && |
2048 | udp_lib_checksum_complete(skb)) | |
e6afc8ac | 2049 | goto csum_error; |
ce25d66a | 2050 | |
ba66bbe5 | 2051 | if (sk_filter_trim_cap(sk, skb, sizeof(struct udphdr))) |
a6127697 | 2052 | goto drop; |
db8dac20 | 2053 | |
e6afc8ac | 2054 | udp_csum_pull_header(skb); |
db8dac20 | 2055 | |
fbf8866d | 2056 | ipv4_pktinfo_prepare(sk, skb); |
850cbadd | 2057 | return __udp_queue_rcv_skb(sk, skb); |
db8dac20 | 2058 | |
6a5dc9e5 | 2059 | csum_error: |
02c22347 | 2060 | __UDP_INC_STATS(sock_net(sk), UDP_MIB_CSUMERRORS, is_udplite); |
db8dac20 | 2061 | drop: |
02c22347 | 2062 | __UDP_INC_STATS(sock_net(sk), UDP_MIB_INERRORS, is_udplite); |
8edf19c2 | 2063 | atomic_inc(&sk->sk_drops); |
db8dac20 DM |
2064 | kfree_skb(skb); |
2065 | return -1; | |
2066 | } | |
2067 | ||
cf329aa4 PA |
2068 | static int udp_queue_rcv_skb(struct sock *sk, struct sk_buff *skb) |
2069 | { | |
2070 | struct sk_buff *next, *segs; | |
2071 | int ret; | |
2072 | ||
2073 | if (likely(!udp_unexpected_gso(sk, skb))) | |
2074 | return udp_queue_rcv_one_skb(sk, skb); | |
2075 | ||
2076 | BUILD_BUG_ON(sizeof(struct udp_skb_cb) > SKB_SGO_CB_OFFSET); | |
2077 | __skb_push(skb, -skb_mac_offset(skb)); | |
2078 | segs = udp_rcv_segment(sk, skb, true); | |
2079 | for (skb = segs; skb; skb = next) { | |
2080 | next = skb->next; | |
2081 | __skb_pull(skb, skb_transport_offset(skb)); | |
2082 | ret = udp_queue_rcv_one_skb(sk, skb); | |
2083 | if (ret > 0) | |
2084 | ip_protocol_deliver_rcu(dev_net(skb->dev), skb, -ret); | |
2085 | } | |
2086 | return 0; | |
2087 | } | |
2088 | ||
97502231 | 2089 | /* For TCP sockets, sk_rx_dst is protected by socket lock |
e47eb5df | 2090 | * For UDP, we use xchg() to guard against concurrent changes. |
97502231 | 2091 | */ |
64f0f5d1 | 2092 | bool udp_sk_rx_dst_set(struct sock *sk, struct dst_entry *dst) |
421b3885 | 2093 | { |
97502231 ED |
2094 | struct dst_entry *old; |
2095 | ||
d24406c8 WW |
2096 | if (dst_hold_safe(dst)) { |
2097 | old = xchg(&sk->sk_rx_dst, dst); | |
2098 | dst_release(old); | |
64f0f5d1 | 2099 | return old != dst; |
d24406c8 | 2100 | } |
64f0f5d1 | 2101 | return false; |
421b3885 | 2102 | } |
c9f2c1ae | 2103 | EXPORT_SYMBOL(udp_sk_rx_dst_set); |
421b3885 | 2104 | |
db8dac20 DM |
2105 | /* |
2106 | * Multicasts and broadcasts go to each listener. | |
2107 | * | |
1240d137 | 2108 | * Note: called only from the BH handler context. |
db8dac20 | 2109 | */ |
e3163493 | 2110 | static int __udp4_lib_mcast_deliver(struct net *net, struct sk_buff *skb, |
db8dac20 DM |
2111 | struct udphdr *uh, |
2112 | __be32 saddr, __be32 daddr, | |
36cbb245 RJ |
2113 | struct udp_table *udptable, |
2114 | int proto) | |
db8dac20 | 2115 | { |
ca065d0c | 2116 | struct sock *sk, *first = NULL; |
5cf3d461 DH |
2117 | unsigned short hnum = ntohs(uh->dest); |
2118 | struct udp_hslot *hslot = udp_hashslot(udptable, net, hnum); | |
2dc41cff | 2119 | unsigned int hash2 = 0, hash2_any = 0, use_hash2 = (hslot->count > 10); |
ca065d0c ED |
2120 | unsigned int offset = offsetof(typeof(*sk), sk_node); |
2121 | int dif = skb->dev->ifindex; | |
fb74c277 | 2122 | int sdif = inet_sdif(skb); |
ca065d0c ED |
2123 | struct hlist_node *node; |
2124 | struct sk_buff *nskb; | |
2dc41cff DH |
2125 | |
2126 | if (use_hash2) { | |
f0b1e64c | 2127 | hash2_any = ipv4_portaddr_hash(net, htonl(INADDR_ANY), hnum) & |
73e2d5e3 | 2128 | udptable->mask; |
f0b1e64c | 2129 | hash2 = ipv4_portaddr_hash(net, daddr, hnum) & udptable->mask; |
2dc41cff | 2130 | start_lookup: |
73e2d5e3 | 2131 | hslot = &udptable->hash2[hash2]; |
2dc41cff DH |
2132 | offset = offsetof(typeof(*sk), __sk_common.skc_portaddr_node); |
2133 | } | |
db8dac20 | 2134 | |
ca065d0c ED |
2135 | sk_for_each_entry_offset_rcu(sk, node, &hslot->head, offset) { |
2136 | if (!__udp_is_mcast_sock(net, sk, uh->dest, daddr, | |
fb74c277 | 2137 | uh->source, saddr, dif, sdif, hnum)) |
ca065d0c ED |
2138 | continue; |
2139 | ||
2140 | if (!first) { | |
2141 | first = sk; | |
2142 | continue; | |
1240d137 | 2143 | } |
ca065d0c | 2144 | nskb = skb_clone(skb, GFP_ATOMIC); |
1240d137 | 2145 | |
ca065d0c ED |
2146 | if (unlikely(!nskb)) { |
2147 | atomic_inc(&sk->sk_drops); | |
02c22347 ED |
2148 | __UDP_INC_STATS(net, UDP_MIB_RCVBUFERRORS, |
2149 | IS_UDPLITE(sk)); | |
2150 | __UDP_INC_STATS(net, UDP_MIB_INERRORS, | |
2151 | IS_UDPLITE(sk)); | |
ca065d0c ED |
2152 | continue; |
2153 | } | |
2154 | if (udp_queue_rcv_skb(sk, nskb) > 0) | |
2155 | consume_skb(nskb); | |
2156 | } | |
1240d137 | 2157 | |
2dc41cff DH |
2158 | /* Also lookup *:port if we are using hash2 and haven't done so yet. */ |
2159 | if (use_hash2 && hash2 != hash2_any) { | |
2160 | hash2 = hash2_any; | |
2161 | goto start_lookup; | |
2162 | } | |
2163 | ||
ca065d0c ED |
2164 | if (first) { |
2165 | if (udp_queue_rcv_skb(first, skb) > 0) | |
2166 | consume_skb(skb); | |
1240d137 | 2167 | } else { |
ca065d0c | 2168 | kfree_skb(skb); |
02c22347 ED |
2169 | __UDP_INC_STATS(net, UDP_MIB_IGNOREDMULTI, |
2170 | proto == IPPROTO_UDPLITE); | |
1240d137 | 2171 | } |
db8dac20 DM |
2172 | return 0; |
2173 | } | |
2174 | ||
2175 | /* Initialize UDP checksum. If exited with zero value (success), | |
2176 | * CHECKSUM_UNNECESSARY means, that no more checks are required. | |
2177 | * Otherwise, csum completion requires chacksumming packet body, | |
2178 | * including udp header and folding it to skb->csum. | |
2179 | */ | |
2180 | static inline int udp4_csum_init(struct sk_buff *skb, struct udphdr *uh, | |
2181 | int proto) | |
2182 | { | |
db8dac20 DM |
2183 | int err; |
2184 | ||
2185 | UDP_SKB_CB(skb)->partial_cov = 0; | |
2186 | UDP_SKB_CB(skb)->cscov = skb->len; | |
2187 | ||
2188 | if (proto == IPPROTO_UDPLITE) { | |
2189 | err = udplite_checksum_init(skb, uh); | |
2190 | if (err) | |
2191 | return err; | |
15f35d49 AK |
2192 | |
2193 | if (UDP_SKB_CB(skb)->partial_cov) { | |
2194 | skb->csum = inet_compute_pseudo(skb, proto); | |
2195 | return 0; | |
2196 | } | |
db8dac20 DM |
2197 | } |
2198 | ||
b46d9f62 HFS |
2199 | /* Note, we are only interested in != 0 or == 0, thus the |
2200 | * force to int. | |
2201 | */ | |
db4f1be3 ST |
2202 | err = (__force int)skb_checksum_init_zero_check(skb, proto, uh->check, |
2203 | inet_compute_pseudo); | |
2204 | if (err) | |
2205 | return err; | |
2206 | ||
2207 | if (skb->ip_summed == CHECKSUM_COMPLETE && !skb->csum_valid) { | |
2208 | /* If SW calculated the value, we know it's bad */ | |
2209 | if (skb->csum_complete_sw) | |
2210 | return 1; | |
2211 | ||
2212 | /* HW says the value is bad. Let's validate that. | |
2213 | * skb->csum is no longer the full packet checksum, | |
2214 | * so don't treat it as such. | |
2215 | */ | |
2216 | skb_checksum_complete_unset(skb); | |
2217 | } | |
2218 | ||
2219 | return 0; | |
db8dac20 DM |
2220 | } |
2221 | ||
2b5a9217 PA |
2222 | /* wrapper for udp_queue_rcv_skb tacking care of csum conversion and |
2223 | * return code conversion for ip layer consumption | |
2224 | */ | |
2225 | static int udp_unicast_rcv_skb(struct sock *sk, struct sk_buff *skb, | |
2226 | struct udphdr *uh) | |
2227 | { | |
2228 | int ret; | |
2229 | ||
2230 | if (inet_get_convert_csum(sk) && uh->check && !IS_UDPLITE(sk)) | |
2231 | skb_checksum_try_convert(skb, IPPROTO_UDP, uh->check, | |
2232 | inet_compute_pseudo); | |
2233 | ||
2234 | ret = udp_queue_rcv_skb(sk, skb); | |
2235 | ||
2236 | /* a return value > 0 means to resubmit the input, but | |
2237 | * it wants the return to be -protocol, or 0 | |
2238 | */ | |
2239 | if (ret > 0) | |
2240 | return -ret; | |
2241 | return 0; | |
2242 | } | |
2243 | ||
db8dac20 DM |
2244 | /* |
2245 | * All we need to do is get the socket, and then do a checksum. | |
2246 | */ | |
2247 | ||
645ca708 | 2248 | int __udp4_lib_rcv(struct sk_buff *skb, struct udp_table *udptable, |
db8dac20 DM |
2249 | int proto) |
2250 | { | |
2251 | struct sock *sk; | |
7b5e56f9 | 2252 | struct udphdr *uh; |
db8dac20 | 2253 | unsigned short ulen; |
adf30907 | 2254 | struct rtable *rt = skb_rtable(skb); |
2783ef23 | 2255 | __be32 saddr, daddr; |
0283328e | 2256 | struct net *net = dev_net(skb->dev); |
db8dac20 DM |
2257 | |
2258 | /* | |
2259 | * Validate the packet. | |
2260 | */ | |
2261 | if (!pskb_may_pull(skb, sizeof(struct udphdr))) | |
2262 | goto drop; /* No space for header. */ | |
2263 | ||
7b5e56f9 | 2264 | uh = udp_hdr(skb); |
db8dac20 | 2265 | ulen = ntohs(uh->len); |
ccc2d97c BM |
2266 | saddr = ip_hdr(skb)->saddr; |
2267 | daddr = ip_hdr(skb)->daddr; | |
2268 | ||
db8dac20 DM |
2269 | if (ulen > skb->len) |
2270 | goto short_packet; | |
2271 | ||
2272 | if (proto == IPPROTO_UDP) { | |
2273 | /* UDP validates ulen. */ | |
2274 | if (ulen < sizeof(*uh) || pskb_trim_rcsum(skb, ulen)) | |
2275 | goto short_packet; | |
2276 | uh = udp_hdr(skb); | |
2277 | } | |
2278 | ||
2279 | if (udp4_csum_init(skb, uh, proto)) | |
2280 | goto csum_error; | |
2281 | ||
8afdd99a ED |
2282 | sk = skb_steal_sock(skb); |
2283 | if (sk) { | |
97502231 | 2284 | struct dst_entry *dst = skb_dst(skb); |
421b3885 | 2285 | int ret; |
421b3885 | 2286 | |
97502231 ED |
2287 | if (unlikely(sk->sk_rx_dst != dst)) |
2288 | udp_sk_rx_dst_set(sk, dst); | |
db8dac20 | 2289 | |
2b5a9217 | 2290 | ret = udp_unicast_rcv_skb(sk, skb, uh); |
8afdd99a | 2291 | sock_put(sk); |
2b5a9217 | 2292 | return ret; |
421b3885 | 2293 | } |
db8dac20 | 2294 | |
c18450a5 FF |
2295 | if (rt->rt_flags & (RTCF_BROADCAST|RTCF_MULTICAST)) |
2296 | return __udp4_lib_mcast_deliver(net, skb, uh, | |
36cbb245 | 2297 | saddr, daddr, udptable, proto); |
c18450a5 FF |
2298 | |
2299 | sk = __udp4_lib_lookup_skb(skb, uh->source, uh->dest, udptable); | |
2b5a9217 PA |
2300 | if (sk) |
2301 | return udp_unicast_rcv_skb(sk, skb, uh); | |
db8dac20 DM |
2302 | |
2303 | if (!xfrm4_policy_check(NULL, XFRM_POLICY_IN, skb)) | |
2304 | goto drop; | |
2305 | nf_reset(skb); | |
2306 | ||
2307 | /* No socket. Drop packet silently, if checksum is wrong */ | |
2308 | if (udp_lib_checksum_complete(skb)) | |
2309 | goto csum_error; | |
2310 | ||
02c22347 | 2311 | __UDP_INC_STATS(net, UDP_MIB_NOPORTS, proto == IPPROTO_UDPLITE); |
db8dac20 DM |
2312 | icmp_send(skb, ICMP_DEST_UNREACH, ICMP_PORT_UNREACH, 0); |
2313 | ||
2314 | /* | |
2315 | * Hmm. We got an UDP packet to a port to which we | |
2316 | * don't wanna listen. Ignore it. | |
2317 | */ | |
2318 | kfree_skb(skb); | |
2319 | return 0; | |
2320 | ||
2321 | short_packet: | |
ba7a46f1 JP |
2322 | net_dbg_ratelimited("UDP%s: short packet: From %pI4:%u %d/%d to %pI4:%u\n", |
2323 | proto == IPPROTO_UDPLITE ? "Lite" : "", | |
2324 | &saddr, ntohs(uh->source), | |
2325 | ulen, skb->len, | |
2326 | &daddr, ntohs(uh->dest)); | |
db8dac20 DM |
2327 | goto drop; |
2328 | ||
2329 | csum_error: | |
2330 | /* | |
2331 | * RFC1122: OK. Discards the bad packet silently (as far as | |
2332 | * the network is concerned, anyway) as per 4.1.3.4 (MUST). | |
2333 | */ | |
ba7a46f1 JP |
2334 | net_dbg_ratelimited("UDP%s: bad checksum. From %pI4:%u to %pI4:%u ulen %d\n", |
2335 | proto == IPPROTO_UDPLITE ? "Lite" : "", | |
2336 | &saddr, ntohs(uh->source), &daddr, ntohs(uh->dest), | |
2337 | ulen); | |
02c22347 | 2338 | __UDP_INC_STATS(net, UDP_MIB_CSUMERRORS, proto == IPPROTO_UDPLITE); |
db8dac20 | 2339 | drop: |
02c22347 | 2340 | __UDP_INC_STATS(net, UDP_MIB_INERRORS, proto == IPPROTO_UDPLITE); |
db8dac20 DM |
2341 | kfree_skb(skb); |
2342 | return 0; | |
2343 | } | |
2344 | ||
421b3885 SB |
2345 | /* We can only early demux multicast if there is a single matching socket. |
2346 | * If more than one socket found returns NULL | |
2347 | */ | |
2348 | static struct sock *__udp4_lib_mcast_demux_lookup(struct net *net, | |
2349 | __be16 loc_port, __be32 loc_addr, | |
2350 | __be16 rmt_port, __be32 rmt_addr, | |
fb74c277 | 2351 | int dif, int sdif) |
421b3885 SB |
2352 | { |
2353 | struct sock *sk, *result; | |
421b3885 | 2354 | unsigned short hnum = ntohs(loc_port); |
ca065d0c | 2355 | unsigned int slot = udp_hashfn(net, hnum, udp_table.mask); |
421b3885 SB |
2356 | struct udp_hslot *hslot = &udp_table.hash[slot]; |
2357 | ||
63c6f81c ED |
2358 | /* Do not bother scanning a too big list */ |
2359 | if (hslot->count > 10) | |
2360 | return NULL; | |
2361 | ||
421b3885 | 2362 | result = NULL; |
ca065d0c ED |
2363 | sk_for_each_rcu(sk, &hslot->head) { |
2364 | if (__udp_is_mcast_sock(net, sk, loc_port, loc_addr, | |
fb74c277 | 2365 | rmt_port, rmt_addr, dif, sdif, hnum)) { |
ca065d0c ED |
2366 | if (result) |
2367 | return NULL; | |
421b3885 | 2368 | result = sk; |
421b3885 SB |
2369 | } |
2370 | } | |
ca065d0c | 2371 | |
421b3885 SB |
2372 | return result; |
2373 | } | |
2374 | ||
2375 | /* For unicast we should only early demux connected sockets or we can | |
2376 | * break forwarding setups. The chains here can be long so only check | |
2377 | * if the first socket is an exact match and if not move on. | |
2378 | */ | |
2379 | static struct sock *__udp4_lib_demux_lookup(struct net *net, | |
2380 | __be16 loc_port, __be32 loc_addr, | |
2381 | __be16 rmt_port, __be32 rmt_addr, | |
3fa6f616 | 2382 | int dif, int sdif) |
421b3885 | 2383 | { |
421b3885 | 2384 | unsigned short hnum = ntohs(loc_port); |
f0b1e64c | 2385 | unsigned int hash2 = ipv4_portaddr_hash(net, loc_addr, hnum); |
421b3885 SB |
2386 | unsigned int slot2 = hash2 & udp_table.mask; |
2387 | struct udp_hslot *hslot2 = &udp_table.hash2[slot2]; | |
c7228317 | 2388 | INET_ADDR_COOKIE(acookie, rmt_addr, loc_addr); |
421b3885 | 2389 | const __portpair ports = INET_COMBINED_PORTS(rmt_port, hnum); |
ca065d0c | 2390 | struct sock *sk; |
421b3885 | 2391 | |
ca065d0c ED |
2392 | udp_portaddr_for_each_entry_rcu(sk, &hslot2->head) { |
2393 | if (INET_MATCH(sk, net, acookie, rmt_addr, | |
3fa6f616 | 2394 | loc_addr, ports, dif, sdif)) |
ca065d0c | 2395 | return sk; |
421b3885 SB |
2396 | /* Only check first socket in chain */ |
2397 | break; | |
2398 | } | |
ca065d0c | 2399 | return NULL; |
421b3885 SB |
2400 | } |
2401 | ||
7487449c | 2402 | int udp_v4_early_demux(struct sk_buff *skb) |
421b3885 | 2403 | { |
610438b7 | 2404 | struct net *net = dev_net(skb->dev); |
bc044e8d | 2405 | struct in_device *in_dev = NULL; |
610438b7 ED |
2406 | const struct iphdr *iph; |
2407 | const struct udphdr *uh; | |
ca065d0c | 2408 | struct sock *sk = NULL; |
421b3885 | 2409 | struct dst_entry *dst; |
421b3885 | 2410 | int dif = skb->dev->ifindex; |
fb74c277 | 2411 | int sdif = inet_sdif(skb); |
6e540309 | 2412 | int ours; |
421b3885 SB |
2413 | |
2414 | /* validate the packet */ | |
2415 | if (!pskb_may_pull(skb, skb_transport_offset(skb) + sizeof(struct udphdr))) | |
7487449c | 2416 | return 0; |
421b3885 | 2417 | |
610438b7 ED |
2418 | iph = ip_hdr(skb); |
2419 | uh = udp_hdr(skb); | |
2420 | ||
996b44fc | 2421 | if (skb->pkt_type == PACKET_MULTICAST) { |
bc044e8d | 2422 | in_dev = __in_dev_get_rcu(skb->dev); |
6e540309 SB |
2423 | |
2424 | if (!in_dev) | |
7487449c | 2425 | return 0; |
6e540309 | 2426 | |
996b44fc PA |
2427 | ours = ip_check_mc_rcu(in_dev, iph->daddr, iph->saddr, |
2428 | iph->protocol); | |
2429 | if (!ours) | |
2430 | return 0; | |
ad0ea198 | 2431 | |
421b3885 | 2432 | sk = __udp4_lib_mcast_demux_lookup(net, uh->dest, iph->daddr, |
fb74c277 DA |
2433 | uh->source, iph->saddr, |
2434 | dif, sdif); | |
6e540309 | 2435 | } else if (skb->pkt_type == PACKET_HOST) { |
421b3885 | 2436 | sk = __udp4_lib_demux_lookup(net, uh->dest, iph->daddr, |
3fa6f616 | 2437 | uh->source, iph->saddr, dif, sdif); |
6e540309 | 2438 | } |
421b3885 | 2439 | |
41c6d650 | 2440 | if (!sk || !refcount_inc_not_zero(&sk->sk_refcnt)) |
7487449c | 2441 | return 0; |
421b3885 SB |
2442 | |
2443 | skb->sk = sk; | |
82eabd9e | 2444 | skb->destructor = sock_efree; |
10e2eb87 | 2445 | dst = READ_ONCE(sk->sk_rx_dst); |
421b3885 SB |
2446 | |
2447 | if (dst) | |
2448 | dst = dst_check(dst, 0); | |
10e2eb87 | 2449 | if (dst) { |
bc044e8d PA |
2450 | u32 itag = 0; |
2451 | ||
d24406c8 WW |
2452 | /* set noref for now. |
2453 | * any place which wants to hold dst has to call | |
2454 | * dst_hold_safe() | |
2455 | */ | |
2456 | skb_dst_set_noref(skb, dst); | |
bc044e8d PA |
2457 | |
2458 | /* for unconnected multicast sockets we need to validate | |
2459 | * the source on each packet | |
2460 | */ | |
2461 | if (!inet_sk(sk)->inet_daddr && in_dev) | |
2462 | return ip_mc_validate_source(skb, iph->daddr, | |
2463 | iph->saddr, iph->tos, | |
2464 | skb->dev, in_dev, &itag); | |
10e2eb87 | 2465 | } |
7487449c | 2466 | return 0; |
421b3885 SB |
2467 | } |
2468 | ||
db8dac20 DM |
2469 | int udp_rcv(struct sk_buff *skb) |
2470 | { | |
645ca708 | 2471 | return __udp4_lib_rcv(skb, &udp_table, IPPROTO_UDP); |
db8dac20 DM |
2472 | } |
2473 | ||
7d06b2e0 | 2474 | void udp_destroy_sock(struct sock *sk) |
db8dac20 | 2475 | { |
44046a59 | 2476 | struct udp_sock *up = udp_sk(sk); |
8a74ad60 | 2477 | bool slow = lock_sock_fast(sk); |
db8dac20 | 2478 | udp_flush_pending_frames(sk); |
8a74ad60 | 2479 | unlock_sock_fast(sk, slow); |
60fb9567 PA |
2480 | if (static_branch_unlikely(&udp_encap_needed_key)) { |
2481 | if (up->encap_type) { | |
2482 | void (*encap_destroy)(struct sock *sk); | |
2483 | encap_destroy = READ_ONCE(up->encap_destroy); | |
2484 | if (encap_destroy) | |
2485 | encap_destroy(sk); | |
2486 | } | |
2487 | if (up->encap_enabled) | |
9c480601 | 2488 | static_branch_dec(&udp_encap_needed_key); |
44046a59 | 2489 | } |
db8dac20 DM |
2490 | } |
2491 | ||
1da177e4 LT |
2492 | /* |
2493 | * Socket option code for UDP | |
2494 | */ | |
4c0a6cb0 | 2495 | int udp_lib_setsockopt(struct sock *sk, int level, int optname, |
b7058842 | 2496 | char __user *optval, unsigned int optlen, |
4c0a6cb0 | 2497 | int (*push_pending_frames)(struct sock *)) |
1da177e4 LT |
2498 | { |
2499 | struct udp_sock *up = udp_sk(sk); | |
1c19448c | 2500 | int val, valbool; |
1da177e4 | 2501 | int err = 0; |
b2bf1e26 | 2502 | int is_udplite = IS_UDPLITE(sk); |
1da177e4 | 2503 | |
c482c568 | 2504 | if (optlen < sizeof(int)) |
1da177e4 LT |
2505 | return -EINVAL; |
2506 | ||
2507 | if (get_user(val, (int __user *)optval)) | |
2508 | return -EFAULT; | |
2509 | ||
1c19448c TH |
2510 | valbool = val ? 1 : 0; |
2511 | ||
6516c655 | 2512 | switch (optname) { |
1da177e4 LT |
2513 | case UDP_CORK: |
2514 | if (val != 0) { | |
2515 | up->corkflag = 1; | |
2516 | } else { | |
2517 | up->corkflag = 0; | |
2518 | lock_sock(sk); | |
4243cdc2 | 2519 | push_pending_frames(sk); |
1da177e4 LT |
2520 | release_sock(sk); |
2521 | } | |
2522 | break; | |
e905a9ed | 2523 | |
1da177e4 LT |
2524 | case UDP_ENCAP: |
2525 | switch (val) { | |
2526 | case 0: | |
2527 | case UDP_ENCAP_ESPINUDP: | |
2528 | case UDP_ENCAP_ESPINUDP_NON_IKE: | |
067b207b JC |
2529 | up->encap_rcv = xfrm4_udp_encap_rcv; |
2530 | /* FALLTHROUGH */ | |
342f0234 | 2531 | case UDP_ENCAP_L2TPINUDP: |
1da177e4 | 2532 | up->encap_type = val; |
60fb9567 PA |
2533 | lock_sock(sk); |
2534 | udp_tunnel_encap_enable(sk->sk_socket); | |
2535 | release_sock(sk); | |
1da177e4 LT |
2536 | break; |
2537 | default: | |
2538 | err = -ENOPROTOOPT; | |
2539 | break; | |
2540 | } | |
2541 | break; | |
2542 | ||
1c19448c TH |
2543 | case UDP_NO_CHECK6_TX: |
2544 | up->no_check6_tx = valbool; | |
2545 | break; | |
2546 | ||
2547 | case UDP_NO_CHECK6_RX: | |
2548 | up->no_check6_rx = valbool; | |
2549 | break; | |
2550 | ||
bec1f6f6 WB |
2551 | case UDP_SEGMENT: |
2552 | if (val < 0 || val > USHRT_MAX) | |
2553 | return -EINVAL; | |
2554 | up->gso_size = val; | |
2555 | break; | |
2556 | ||
e20cf8d3 PA |
2557 | case UDP_GRO: |
2558 | lock_sock(sk); | |
2559 | if (valbool) | |
2560 | udp_tunnel_encap_enable(sk->sk_socket); | |
2561 | up->gro_enabled = valbool; | |
2562 | release_sock(sk); | |
2563 | break; | |
2564 | ||
ba4e58ec GR |
2565 | /* |
2566 | * UDP-Lite's partial checksum coverage (RFC 3828). | |
2567 | */ | |
2568 | /* The sender sets actual checksum coverage length via this option. | |
2569 | * The case coverage > packet length is handled by send module. */ | |
2570 | case UDPLITE_SEND_CSCOV: | |
b2bf1e26 | 2571 | if (!is_udplite) /* Disable the option on UDP sockets */ |
ba4e58ec GR |
2572 | return -ENOPROTOOPT; |
2573 | if (val != 0 && val < 8) /* Illegal coverage: use default (8) */ | |
2574 | val = 8; | |
4be929be AD |
2575 | else if (val > USHRT_MAX) |
2576 | val = USHRT_MAX; | |
ba4e58ec GR |
2577 | up->pcslen = val; |
2578 | up->pcflag |= UDPLITE_SEND_CC; | |
2579 | break; | |
2580 | ||
e905a9ed YH |
2581 | /* The receiver specifies a minimum checksum coverage value. To make |
2582 | * sense, this should be set to at least 8 (as done below). If zero is | |
ba4e58ec GR |
2583 | * used, this again means full checksum coverage. */ |
2584 | case UDPLITE_RECV_CSCOV: | |
b2bf1e26 | 2585 | if (!is_udplite) /* Disable the option on UDP sockets */ |
ba4e58ec GR |
2586 | return -ENOPROTOOPT; |
2587 | if (val != 0 && val < 8) /* Avoid silly minimal values. */ | |
2588 | val = 8; | |
4be929be AD |
2589 | else if (val > USHRT_MAX) |
2590 | val = USHRT_MAX; | |
ba4e58ec GR |
2591 | up->pcrlen = val; |
2592 | up->pcflag |= UDPLITE_RECV_CC; | |
2593 | break; | |
2594 | ||
1da177e4 LT |
2595 | default: |
2596 | err = -ENOPROTOOPT; | |
2597 | break; | |
6516c655 | 2598 | } |
1da177e4 LT |
2599 | |
2600 | return err; | |
2601 | } | |
c482c568 | 2602 | EXPORT_SYMBOL(udp_lib_setsockopt); |
1da177e4 | 2603 | |
db8dac20 | 2604 | int udp_setsockopt(struct sock *sk, int level, int optname, |
b7058842 | 2605 | char __user *optval, unsigned int optlen) |
db8dac20 DM |
2606 | { |
2607 | if (level == SOL_UDP || level == SOL_UDPLITE) | |
2608 | return udp_lib_setsockopt(sk, level, optname, optval, optlen, | |
2609 | udp_push_pending_frames); | |
2610 | return ip_setsockopt(sk, level, optname, optval, optlen); | |
2611 | } | |
2612 | ||
2613 | #ifdef CONFIG_COMPAT | |
2614 | int compat_udp_setsockopt(struct sock *sk, int level, int optname, | |
b7058842 | 2615 | char __user *optval, unsigned int optlen) |
db8dac20 DM |
2616 | { |
2617 | if (level == SOL_UDP || level == SOL_UDPLITE) | |
2618 | return udp_lib_setsockopt(sk, level, optname, optval, optlen, | |
2619 | udp_push_pending_frames); | |
2620 | return compat_ip_setsockopt(sk, level, optname, optval, optlen); | |
2621 | } | |
2622 | #endif | |
2623 | ||
4c0a6cb0 GR |
2624 | int udp_lib_getsockopt(struct sock *sk, int level, int optname, |
2625 | char __user *optval, int __user *optlen) | |
1da177e4 LT |
2626 | { |
2627 | struct udp_sock *up = udp_sk(sk); | |
2628 | int val, len; | |
2629 | ||
c482c568 | 2630 | if (get_user(len, optlen)) |
1da177e4 LT |
2631 | return -EFAULT; |
2632 | ||
2633 | len = min_t(unsigned int, len, sizeof(int)); | |
e905a9ed | 2634 | |
6516c655 | 2635 | if (len < 0) |
1da177e4 LT |
2636 | return -EINVAL; |
2637 | ||
6516c655 | 2638 | switch (optname) { |
1da177e4 LT |
2639 | case UDP_CORK: |
2640 | val = up->corkflag; | |
2641 | break; | |
2642 | ||
2643 | case UDP_ENCAP: | |
2644 | val = up->encap_type; | |
2645 | break; | |
2646 | ||
1c19448c TH |
2647 | case UDP_NO_CHECK6_TX: |
2648 | val = up->no_check6_tx; | |
2649 | break; | |
2650 | ||
2651 | case UDP_NO_CHECK6_RX: | |
2652 | val = up->no_check6_rx; | |
2653 | break; | |
2654 | ||
bec1f6f6 WB |
2655 | case UDP_SEGMENT: |
2656 | val = up->gso_size; | |
2657 | break; | |
2658 | ||
ba4e58ec GR |
2659 | /* The following two cannot be changed on UDP sockets, the return is |
2660 | * always 0 (which corresponds to the full checksum coverage of UDP). */ | |
2661 | case UDPLITE_SEND_CSCOV: | |
2662 | val = up->pcslen; | |
2663 | break; | |
2664 | ||
2665 | case UDPLITE_RECV_CSCOV: | |
2666 | val = up->pcrlen; | |
2667 | break; | |
2668 | ||
1da177e4 LT |
2669 | default: |
2670 | return -ENOPROTOOPT; | |
6516c655 | 2671 | } |
1da177e4 | 2672 | |
6516c655 | 2673 | if (put_user(len, optlen)) |
e905a9ed | 2674 | return -EFAULT; |
c482c568 | 2675 | if (copy_to_user(optval, &val, len)) |
1da177e4 | 2676 | return -EFAULT; |
e905a9ed | 2677 | return 0; |
1da177e4 | 2678 | } |
c482c568 | 2679 | EXPORT_SYMBOL(udp_lib_getsockopt); |
1da177e4 | 2680 | |
db8dac20 DM |
2681 | int udp_getsockopt(struct sock *sk, int level, int optname, |
2682 | char __user *optval, int __user *optlen) | |
2683 | { | |
2684 | if (level == SOL_UDP || level == SOL_UDPLITE) | |
2685 | return udp_lib_getsockopt(sk, level, optname, optval, optlen); | |
2686 | return ip_getsockopt(sk, level, optname, optval, optlen); | |
2687 | } | |
2688 | ||
2689 | #ifdef CONFIG_COMPAT | |
2690 | int compat_udp_getsockopt(struct sock *sk, int level, int optname, | |
2691 | char __user *optval, int __user *optlen) | |
2692 | { | |
2693 | if (level == SOL_UDP || level == SOL_UDPLITE) | |
2694 | return udp_lib_getsockopt(sk, level, optname, optval, optlen); | |
2695 | return compat_ip_getsockopt(sk, level, optname, optval, optlen); | |
2696 | } | |
2697 | #endif | |
1da177e4 LT |
2698 | /** |
2699 | * udp_poll - wait for a UDP event. | |
2700 | * @file - file struct | |
2701 | * @sock - socket | |
a11e1d43 | 2702 | * @wait - poll table |
1da177e4 | 2703 | * |
e905a9ed | 2704 | * This is same as datagram poll, except for the special case of |
1da177e4 LT |
2705 | * blocking sockets. If application is using a blocking fd |
2706 | * and a packet with checksum error is in the queue; | |
2707 | * then it could get return from select indicating data available | |
2708 | * but then block when reading it. Add special case code | |
2709 | * to work around these arguably broken applications. | |
2710 | */ | |
a11e1d43 | 2711 | __poll_t udp_poll(struct file *file, struct socket *sock, poll_table *wait) |
1da177e4 | 2712 | { |
a11e1d43 | 2713 | __poll_t mask = datagram_poll(file, sock, wait); |
1da177e4 | 2714 | struct sock *sk = sock->sk; |
ba4e58ec | 2715 | |
2276f58a | 2716 | if (!skb_queue_empty(&udp_sk(sk)->reader_queue)) |
a9a08845 | 2717 | mask |= EPOLLIN | EPOLLRDNORM; |
2276f58a | 2718 | |
1da177e4 | 2719 | /* Check for false positives due to checksum errors */ |
a11e1d43 | 2720 | if ((mask & EPOLLRDNORM) && !(file->f_flags & O_NONBLOCK) && |
e83c6744 | 2721 | !(sk->sk_shutdown & RCV_SHUTDOWN) && first_packet_length(sk) == -1) |
a9a08845 | 2722 | mask &= ~(EPOLLIN | EPOLLRDNORM); |
1da177e4 LT |
2723 | |
2724 | return mask; | |
e905a9ed | 2725 | |
1da177e4 | 2726 | } |
a11e1d43 | 2727 | EXPORT_SYMBOL(udp_poll); |
1da177e4 | 2728 | |
5d77dca8 DA |
2729 | int udp_abort(struct sock *sk, int err) |
2730 | { | |
2731 | lock_sock(sk); | |
2732 | ||
2733 | sk->sk_err = err; | |
2734 | sk->sk_error_report(sk); | |
286c72de | 2735 | __udp_disconnect(sk, 0); |
5d77dca8 DA |
2736 | |
2737 | release_sock(sk); | |
2738 | ||
2739 | return 0; | |
2740 | } | |
2741 | EXPORT_SYMBOL_GPL(udp_abort); | |
2742 | ||
db8dac20 | 2743 | struct proto udp_prot = { |
1e802951 TZ |
2744 | .name = "UDP", |
2745 | .owner = THIS_MODULE, | |
2746 | .close = udp_lib_close, | |
d74bad4e | 2747 | .pre_connect = udp_pre_connect, |
1e802951 TZ |
2748 | .connect = ip4_datagram_connect, |
2749 | .disconnect = udp_disconnect, | |
2750 | .ioctl = udp_ioctl, | |
2751 | .init = udp_init_sock, | |
2752 | .destroy = udp_destroy_sock, | |
2753 | .setsockopt = udp_setsockopt, | |
2754 | .getsockopt = udp_getsockopt, | |
2755 | .sendmsg = udp_sendmsg, | |
2756 | .recvmsg = udp_recvmsg, | |
2757 | .sendpage = udp_sendpage, | |
2758 | .release_cb = ip4_datagram_release_cb, | |
2759 | .hash = udp_lib_hash, | |
2760 | .unhash = udp_lib_unhash, | |
2761 | .rehash = udp_v4_rehash, | |
2762 | .get_port = udp_v4_get_port, | |
2763 | .memory_allocated = &udp_memory_allocated, | |
2764 | .sysctl_mem = sysctl_udp_mem, | |
2765 | .sysctl_wmem_offset = offsetof(struct net, ipv4.sysctl_udp_wmem_min), | |
2766 | .sysctl_rmem_offset = offsetof(struct net, ipv4.sysctl_udp_rmem_min), | |
2767 | .obj_size = sizeof(struct udp_sock), | |
2768 | .h.udp_table = &udp_table, | |
db8dac20 | 2769 | #ifdef CONFIG_COMPAT |
1e802951 TZ |
2770 | .compat_setsockopt = compat_udp_setsockopt, |
2771 | .compat_getsockopt = compat_udp_getsockopt, | |
db8dac20 | 2772 | #endif |
1e802951 | 2773 | .diag_destroy = udp_abort, |
db8dac20 | 2774 | }; |
c482c568 | 2775 | EXPORT_SYMBOL(udp_prot); |
1da177e4 LT |
2776 | |
2777 | /* ------------------------------------------------------------------------ */ | |
2778 | #ifdef CONFIG_PROC_FS | |
2779 | ||
645ca708 | 2780 | static struct sock *udp_get_first(struct seq_file *seq, int start) |
1da177e4 LT |
2781 | { |
2782 | struct sock *sk; | |
a3d2599b | 2783 | struct udp_seq_afinfo *afinfo = PDE_DATA(file_inode(seq->file)); |
1da177e4 | 2784 | struct udp_iter_state *state = seq->private; |
6f191efe | 2785 | struct net *net = seq_file_net(seq); |
1da177e4 | 2786 | |
a3d2599b | 2787 | for (state->bucket = start; state->bucket <= afinfo->udp_table->mask; |
f86dcc5a | 2788 | ++state->bucket) { |
a3d2599b | 2789 | struct udp_hslot *hslot = &afinfo->udp_table->hash[state->bucket]; |
f86dcc5a | 2790 | |
ca065d0c | 2791 | if (hlist_empty(&hslot->head)) |
f86dcc5a ED |
2792 | continue; |
2793 | ||
645ca708 | 2794 | spin_lock_bh(&hslot->lock); |
ca065d0c | 2795 | sk_for_each(sk, &hslot->head) { |
878628fb | 2796 | if (!net_eq(sock_net(sk), net)) |
a91275ef | 2797 | continue; |
a3d2599b | 2798 | if (sk->sk_family == afinfo->family) |
1da177e4 LT |
2799 | goto found; |
2800 | } | |
645ca708 | 2801 | spin_unlock_bh(&hslot->lock); |
1da177e4 LT |
2802 | } |
2803 | sk = NULL; | |
2804 | found: | |
2805 | return sk; | |
2806 | } | |
2807 | ||
2808 | static struct sock *udp_get_next(struct seq_file *seq, struct sock *sk) | |
2809 | { | |
a3d2599b | 2810 | struct udp_seq_afinfo *afinfo = PDE_DATA(file_inode(seq->file)); |
1da177e4 | 2811 | struct udp_iter_state *state = seq->private; |
6f191efe | 2812 | struct net *net = seq_file_net(seq); |
1da177e4 LT |
2813 | |
2814 | do { | |
ca065d0c | 2815 | sk = sk_next(sk); |
a3d2599b | 2816 | } while (sk && (!net_eq(sock_net(sk), net) || sk->sk_family != afinfo->family)); |
1da177e4 | 2817 | |
645ca708 | 2818 | if (!sk) { |
a3d2599b CH |
2819 | if (state->bucket <= afinfo->udp_table->mask) |
2820 | spin_unlock_bh(&afinfo->udp_table->hash[state->bucket].lock); | |
645ca708 | 2821 | return udp_get_first(seq, state->bucket + 1); |
1da177e4 LT |
2822 | } |
2823 | return sk; | |
2824 | } | |
2825 | ||
2826 | static struct sock *udp_get_idx(struct seq_file *seq, loff_t pos) | |
2827 | { | |
645ca708 | 2828 | struct sock *sk = udp_get_first(seq, 0); |
1da177e4 LT |
2829 | |
2830 | if (sk) | |
6516c655 | 2831 | while (pos && (sk = udp_get_next(seq, sk)) != NULL) |
1da177e4 LT |
2832 | --pos; |
2833 | return pos ? NULL : sk; | |
2834 | } | |
2835 | ||
a3d2599b | 2836 | void *udp_seq_start(struct seq_file *seq, loff_t *pos) |
1da177e4 | 2837 | { |
30842f29 | 2838 | struct udp_iter_state *state = seq->private; |
f86dcc5a | 2839 | state->bucket = MAX_UDP_PORTS; |
30842f29 | 2840 | |
b50660f1 | 2841 | return *pos ? udp_get_idx(seq, *pos-1) : SEQ_START_TOKEN; |
1da177e4 | 2842 | } |
a3d2599b | 2843 | EXPORT_SYMBOL(udp_seq_start); |
1da177e4 | 2844 | |
a3d2599b | 2845 | void *udp_seq_next(struct seq_file *seq, void *v, loff_t *pos) |
1da177e4 LT |
2846 | { |
2847 | struct sock *sk; | |
2848 | ||
b50660f1 | 2849 | if (v == SEQ_START_TOKEN) |
1da177e4 LT |
2850 | sk = udp_get_idx(seq, 0); |
2851 | else | |
2852 | sk = udp_get_next(seq, v); | |
2853 | ||
2854 | ++*pos; | |
2855 | return sk; | |
2856 | } | |
a3d2599b | 2857 | EXPORT_SYMBOL(udp_seq_next); |
1da177e4 | 2858 | |
a3d2599b | 2859 | void udp_seq_stop(struct seq_file *seq, void *v) |
1da177e4 | 2860 | { |
a3d2599b | 2861 | struct udp_seq_afinfo *afinfo = PDE_DATA(file_inode(seq->file)); |
645ca708 ED |
2862 | struct udp_iter_state *state = seq->private; |
2863 | ||
a3d2599b CH |
2864 | if (state->bucket <= afinfo->udp_table->mask) |
2865 | spin_unlock_bh(&afinfo->udp_table->hash[state->bucket].lock); | |
1da177e4 | 2866 | } |
a3d2599b | 2867 | EXPORT_SYMBOL(udp_seq_stop); |
db8dac20 DM |
2868 | |
2869 | /* ------------------------------------------------------------------------ */ | |
5e659e4c | 2870 | static void udp4_format_sock(struct sock *sp, struct seq_file *f, |
652586df | 2871 | int bucket) |
db8dac20 DM |
2872 | { |
2873 | struct inet_sock *inet = inet_sk(sp); | |
c720c7e8 ED |
2874 | __be32 dest = inet->inet_daddr; |
2875 | __be32 src = inet->inet_rcv_saddr; | |
2876 | __u16 destp = ntohs(inet->inet_dport); | |
2877 | __u16 srcp = ntohs(inet->inet_sport); | |
db8dac20 | 2878 | |
f86dcc5a | 2879 | seq_printf(f, "%5d: %08X:%04X %08X:%04X" |
ea9a0379 | 2880 | " %02X %08X:%08X %02X:%08lX %08X %5u %8d %lu %d %pK %u", |
db8dac20 | 2881 | bucket, src, srcp, dest, destp, sp->sk_state, |
31e6d363 | 2882 | sk_wmem_alloc_get(sp), |
6c206b20 | 2883 | udp_rqueue_get(sp), |
a7cb5a49 EB |
2884 | 0, 0L, 0, |
2885 | from_kuid_munged(seq_user_ns(f), sock_i_uid(sp)), | |
2886 | 0, sock_i_ino(sp), | |
41c6d650 | 2887 | refcount_read(&sp->sk_refcnt), sp, |
652586df | 2888 | atomic_read(&sp->sk_drops)); |
db8dac20 DM |
2889 | } |
2890 | ||
2891 | int udp4_seq_show(struct seq_file *seq, void *v) | |
2892 | { | |
652586df | 2893 | seq_setwidth(seq, 127); |
db8dac20 | 2894 | if (v == SEQ_START_TOKEN) |
652586df | 2895 | seq_puts(seq, " sl local_address rem_address st tx_queue " |
db8dac20 | 2896 | "rx_queue tr tm->when retrnsmt uid timeout " |
cb61cb9b | 2897 | "inode ref pointer drops"); |
db8dac20 | 2898 | else { |
db8dac20 DM |
2899 | struct udp_iter_state *state = seq->private; |
2900 | ||
652586df | 2901 | udp4_format_sock(v, seq, state->bucket); |
db8dac20 | 2902 | } |
652586df | 2903 | seq_pad(seq, '\n'); |
db8dac20 DM |
2904 | return 0; |
2905 | } | |
2906 | ||
c3506372 | 2907 | const struct seq_operations udp_seq_ops = { |
a3d2599b CH |
2908 | .start = udp_seq_start, |
2909 | .next = udp_seq_next, | |
2910 | .stop = udp_seq_stop, | |
2911 | .show = udp4_seq_show, | |
2912 | }; | |
c3506372 | 2913 | EXPORT_SYMBOL(udp_seq_ops); |
73cb88ec | 2914 | |
db8dac20 | 2915 | static struct udp_seq_afinfo udp4_seq_afinfo = { |
db8dac20 | 2916 | .family = AF_INET, |
645ca708 | 2917 | .udp_table = &udp_table, |
db8dac20 DM |
2918 | }; |
2919 | ||
2c8c1e72 | 2920 | static int __net_init udp4_proc_init_net(struct net *net) |
15439feb | 2921 | { |
c3506372 CH |
2922 | if (!proc_create_net_data("udp", 0444, net->proc_net, &udp_seq_ops, |
2923 | sizeof(struct udp_iter_state), &udp4_seq_afinfo)) | |
a3d2599b CH |
2924 | return -ENOMEM; |
2925 | return 0; | |
15439feb PE |
2926 | } |
2927 | ||
2c8c1e72 | 2928 | static void __net_exit udp4_proc_exit_net(struct net *net) |
15439feb | 2929 | { |
a3d2599b | 2930 | remove_proc_entry("udp", net->proc_net); |
15439feb PE |
2931 | } |
2932 | ||
2933 | static struct pernet_operations udp4_net_ops = { | |
2934 | .init = udp4_proc_init_net, | |
2935 | .exit = udp4_proc_exit_net, | |
2936 | }; | |
2937 | ||
db8dac20 DM |
2938 | int __init udp4_proc_init(void) |
2939 | { | |
15439feb | 2940 | return register_pernet_subsys(&udp4_net_ops); |
db8dac20 DM |
2941 | } |
2942 | ||
2943 | void udp4_proc_exit(void) | |
2944 | { | |
15439feb | 2945 | unregister_pernet_subsys(&udp4_net_ops); |
db8dac20 | 2946 | } |
1da177e4 LT |
2947 | #endif /* CONFIG_PROC_FS */ |
2948 | ||
f86dcc5a ED |
2949 | static __initdata unsigned long uhash_entries; |
2950 | static int __init set_uhash_entries(char *str) | |
645ca708 | 2951 | { |
413c27d8 EZ |
2952 | ssize_t ret; |
2953 | ||
f86dcc5a ED |
2954 | if (!str) |
2955 | return 0; | |
413c27d8 EZ |
2956 | |
2957 | ret = kstrtoul(str, 0, &uhash_entries); | |
2958 | if (ret) | |
2959 | return 0; | |
2960 | ||
f86dcc5a ED |
2961 | if (uhash_entries && uhash_entries < UDP_HTABLE_SIZE_MIN) |
2962 | uhash_entries = UDP_HTABLE_SIZE_MIN; | |
2963 | return 1; | |
2964 | } | |
2965 | __setup("uhash_entries=", set_uhash_entries); | |
645ca708 | 2966 | |
f86dcc5a ED |
2967 | void __init udp_table_init(struct udp_table *table, const char *name) |
2968 | { | |
2969 | unsigned int i; | |
2970 | ||
31fe62b9 TB |
2971 | table->hash = alloc_large_system_hash(name, |
2972 | 2 * sizeof(struct udp_hslot), | |
2973 | uhash_entries, | |
2974 | 21, /* one slot per 2 MB */ | |
2975 | 0, | |
2976 | &table->log, | |
2977 | &table->mask, | |
2978 | UDP_HTABLE_SIZE_MIN, | |
2979 | 64 * 1024); | |
2980 | ||
512615b6 | 2981 | table->hash2 = table->hash + (table->mask + 1); |
f86dcc5a | 2982 | for (i = 0; i <= table->mask; i++) { |
ca065d0c | 2983 | INIT_HLIST_HEAD(&table->hash[i].head); |
fdcc8aa9 | 2984 | table->hash[i].count = 0; |
645ca708 ED |
2985 | spin_lock_init(&table->hash[i].lock); |
2986 | } | |
512615b6 | 2987 | for (i = 0; i <= table->mask; i++) { |
ca065d0c | 2988 | INIT_HLIST_HEAD(&table->hash2[i].head); |
512615b6 ED |
2989 | table->hash2[i].count = 0; |
2990 | spin_lock_init(&table->hash2[i].lock); | |
2991 | } | |
645ca708 ED |
2992 | } |
2993 | ||
723b8e46 TH |
2994 | u32 udp_flow_hashrnd(void) |
2995 | { | |
2996 | static u32 hashrnd __read_mostly; | |
2997 | ||
2998 | net_get_random_once(&hashrnd, sizeof(hashrnd)); | |
2999 | ||
3000 | return hashrnd; | |
3001 | } | |
3002 | EXPORT_SYMBOL(udp_flow_hashrnd); | |
3003 | ||
1e802951 TZ |
3004 | static void __udp_sysctl_init(struct net *net) |
3005 | { | |
3006 | net->ipv4.sysctl_udp_rmem_min = SK_MEM_QUANTUM; | |
3007 | net->ipv4.sysctl_udp_wmem_min = SK_MEM_QUANTUM; | |
3008 | ||
3009 | #ifdef CONFIG_NET_L3_MASTER_DEV | |
3010 | net->ipv4.sysctl_udp_l3mdev_accept = 0; | |
3011 | #endif | |
3012 | } | |
3013 | ||
3014 | static int __net_init udp_sysctl_init(struct net *net) | |
3015 | { | |
3016 | __udp_sysctl_init(net); | |
3017 | return 0; | |
3018 | } | |
3019 | ||
3020 | static struct pernet_operations __net_initdata udp_sysctl_ops = { | |
fc18999e | 3021 | .init = udp_sysctl_init, |
1e802951 TZ |
3022 | }; |
3023 | ||
95766fff HA |
3024 | void __init udp_init(void) |
3025 | { | |
f03d78db | 3026 | unsigned long limit; |
4b272750 | 3027 | unsigned int i; |
95766fff | 3028 | |
f86dcc5a | 3029 | udp_table_init(&udp_table, "UDP"); |
f03d78db | 3030 | limit = nr_free_buffer_pages() / 8; |
95766fff HA |
3031 | limit = max(limit, 128UL); |
3032 | sysctl_udp_mem[0] = limit / 4 * 3; | |
3033 | sysctl_udp_mem[1] = limit; | |
3034 | sysctl_udp_mem[2] = sysctl_udp_mem[0] * 2; | |
3035 | ||
1e802951 | 3036 | __udp_sysctl_init(&init_net); |
4b272750 ED |
3037 | |
3038 | /* 16 spinlocks per cpu */ | |
3039 | udp_busylocks_log = ilog2(nr_cpu_ids) + 4; | |
3040 | udp_busylocks = kmalloc(sizeof(spinlock_t) << udp_busylocks_log, | |
3041 | GFP_KERNEL); | |
3042 | if (!udp_busylocks) | |
3043 | panic("UDP: failed to alloc udp_busylocks\n"); | |
3044 | for (i = 0; i < (1U << udp_busylocks_log); i++) | |
3045 | spin_lock_init(udp_busylocks + i); | |
1e802951 TZ |
3046 | |
3047 | if (register_pernet_subsys(&udp_sysctl_ops)) | |
3048 | panic("UDP: failed to init sysctl parameters.\n"); | |
95766fff | 3049 | } |